Skip to content

Instantly share code, notes, and snippets.

@joleuger
Last active June 1, 2026 21:29
Show Gist options
  • Select an option

  • Save joleuger/36a1769344a528c185801350acce457e to your computer and use it in GitHub Desktop.

Select an option

Save joleuger/36a1769344a528c185801350acce457e to your computer and use it in GitHub Desktop.
SSE Proxy test

ccache not found, please consider installing for faster builds### ctest_debug

Runs ctest in debug mode

  • status: 0
+ ctest -C Debug --output-on-failure -L main -E 'test-opt|test-backend-ops|test-llama-archs'
Test project /root/llama.me/build-ci-debug
      Start  1: test-tokenizer-0-bert-bge
 1/42 Test  #1: test-tokenizer-0-bert-bge .........   Passed    0.11 sec
      Start  2: test-tokenizer-0-command-r
 2/42 Test  #2: test-tokenizer-0-command-r ........   Passed    1.64 sec
      Start  3: test-tokenizer-0-deepseek-coder
 3/42 Test  #3: test-tokenizer-0-deepseek-coder ...   Passed    0.26 sec
      Start  4: test-tokenizer-0-deepseek-llm
 4/42 Test  #4: test-tokenizer-0-deepseek-llm .....   Passed    0.64 sec
      Start  5: test-tokenizer-0-falcon
 5/42 Test  #5: test-tokenizer-0-falcon ...........   Passed    0.40 sec
      Start  6: test-tokenizer-0-gemma-4
 6/42 Test  #6: test-tokenizer-0-gemma-4 ..........   Passed    1.86 sec
      Start  7: test-tokenizer-0-gpt-2
 7/42 Test  #7: test-tokenizer-0-gpt-2 ............   Passed    0.31 sec
      Start  8: test-tokenizer-0-llama-bpe
 8/42 Test  #8: test-tokenizer-0-llama-bpe ........   Passed    1.02 sec
      Start  9: test-tokenizer-0-llama-spm
 9/42 Test  #9: test-tokenizer-0-llama-spm ........   Passed    0.11 sec
      Start 10: test-tokenizer-0-mpt
10/42 Test #10: test-tokenizer-0-mpt ..............   Passed    0.31 sec
      Start 11: test-tokenizer-0-phi-3
11/42 Test #11: test-tokenizer-0-phi-3 ............   Passed    0.11 sec
      Start 12: test-tokenizer-0-qwen2
12/42 Test #12: test-tokenizer-0-qwen2 ............   Passed    0.89 sec
      Start 13: test-tokenizer-0-qwen35
13/42 Test #13: test-tokenizer-0-qwen35 ...........   Passed    0.89 sec
      Start 14: test-tokenizer-0-refact
14/42 Test #14: test-tokenizer-0-refact ...........   Passed    0.30 sec
      Start 15: test-tokenizer-0-starcoder
15/42 Test #15: test-tokenizer-0-starcoder ........   Passed    0.31 sec
      Start 16: test-tokenizers-ggml-vocabs
16/42 Test #16: test-tokenizers-ggml-vocabs .......   Passed    0.50 sec
      Start 17: test-sampling
17/42 Test #17: test-sampling .....................   Passed    3.86 sec
      Start 18: test-reasoning-budget
18/42 Test #18: test-reasoning-budget .............   Passed    0.01 sec
      Start 19: test-grammar-parser
19/42 Test #19: test-grammar-parser ...............   Passed    0.00 sec
      Start 20: test-grammar-integration
20/42 Test #20: test-grammar-integration ..........   Passed    0.07 sec
      Start 21: test-llama-grammar
21/42 Test #21: test-llama-grammar ................   Passed    0.01 sec
      Start 22: test-chat
22/42 Test #22: test-chat .........................   Passed   19.56 sec
      Start 23: test-json-schema-to-grammar
23/42 Test #23: test-json-schema-to-grammar .......   Passed    2.03 sec
      Start 24: test-tokenizer-1-llama-spm
24/42 Test #24: test-tokenizer-1-llama-spm ........   Passed    0.84 sec
      Start 25: test-chat-peg-parser
25/42 Test #25: test-chat-peg-parser ..............   Passed    0.24 sec
      Start 26: test-jinja
26/42 Test #26: test-jinja ........................   Passed    0.51 sec
      Start 28: test-chat-auto-parser
27/42 Test #28: test-chat-auto-parser .............   Passed    2.51 sec
      Start 29: test-chat-template
28/42 Test #29: test-chat-template ................   Passed    0.51 sec
      Start 30: test-json-partial
29/42 Test #30: test-json-partial .................   Passed    0.01 sec
      Start 31: test-log
30/42 Test #31: test-log ..........................   Passed    0.02 sec
      Start 32: test-peg-parser
31/42 Test #32: test-peg-parser ...................   Passed    0.10 sec
      Start 33: test-regex-partial
32/42 Test #33: test-regex-partial ................   Passed    0.01 sec
      Start 34: test-download-model
33/42 Test #34: test-download-model ...............   Passed    1.97 sec
      Start 35: test-thread-safety
34/42 Test #35: test-thread-safety ................   Passed    1.67 sec
      Start 36: test-arg-parser
35/42 Test #36: test-arg-parser ...................   Passed    0.15 sec
      Start 37: test-gguf
36/42 Test #37: test-gguf .........................   Passed    0.21 sec
      Start 44: test-barrier
37/42 Test #44: test-barrier ......................   Passed    2.00 sec
      Start 45: test-quantize-fns
38/42 Test #45: test-quantize-fns .................   Passed    1.87 sec
      Start 46: test-quantize-perf
39/42 Test #46: test-quantize-perf ................   Passed    0.24 sec
      Start 47: test-rope
40/42 Test #47: test-rope .........................   Passed    0.06 sec
      Start 48: test-mtmd-c-api
41/42 Test #48: test-mtmd-c-api ...................   Passed    0.01 sec
      Start 49: test-alloc
42/42 Test #49: test-alloc ........................   Passed    0.00 sec

100% tests passed, 0 tests failed out of 42

Label Time Summary:
main    =  46.18 sec*proc (41 tests)

Total Test time (real) =  48.17 sec

real	0m48,178s
user	1m45,190s
sys	0m1,528s

ccache not found, please consider installing for faster builds### ctest_release

Runs ctest in release mode

  • status: 0
+ ctest -C Release --output-on-failure -L 'main|python'
Test project /root/llama.me/build-ci-release
      Start  1: test-tokenizer-0-bert-bge
 1/45 Test  #1: test-tokenizer-0-bert-bge .........   Passed    0.03 sec
      Start  2: test-tokenizer-0-command-r
 2/45 Test  #2: test-tokenizer-0-command-r ........   Passed    0.43 sec
      Start  3: test-tokenizer-0-deepseek-coder
 3/45 Test  #3: test-tokenizer-0-deepseek-coder ...   Passed    0.05 sec
      Start  4: test-tokenizer-0-deepseek-llm
 4/45 Test  #4: test-tokenizer-0-deepseek-llm .....   Passed    0.14 sec
      Start  5: test-tokenizer-0-falcon
 5/45 Test  #5: test-tokenizer-0-falcon ...........   Passed    0.08 sec
      Start  6: test-tokenizer-0-gemma-4
 6/45 Test  #6: test-tokenizer-0-gemma-4 ..........   Passed    0.57 sec
      Start  7: test-tokenizer-0-gpt-2
 7/45 Test  #7: test-tokenizer-0-gpt-2 ............   Passed    0.06 sec
      Start  8: test-tokenizer-0-llama-bpe
 8/45 Test  #8: test-tokenizer-0-llama-bpe ........   Passed    0.25 sec
      Start  9: test-tokenizer-0-llama-spm
 9/45 Test  #9: test-tokenizer-0-llama-spm ........   Passed    0.03 sec
      Start 10: test-tokenizer-0-mpt
10/45 Test #10: test-tokenizer-0-mpt ..............   Passed    0.06 sec
      Start 11: test-tokenizer-0-phi-3
11/45 Test #11: test-tokenizer-0-phi-3 ............   Passed    0.03 sec
      Start 12: test-tokenizer-0-qwen2
12/45 Test #12: test-tokenizer-0-qwen2 ............   Passed    0.20 sec
      Start 13: test-tokenizer-0-qwen35
13/45 Test #13: test-tokenizer-0-qwen35 ...........   Passed    0.19 sec
      Start 14: test-tokenizer-0-refact
14/45 Test #14: test-tokenizer-0-refact ...........   Passed    0.06 sec
      Start 15: test-tokenizer-0-starcoder
15/45 Test #15: test-tokenizer-0-starcoder ........   Passed    0.06 sec
      Start 16: test-tokenizers-ggml-vocabs
16/45 Test #16: test-tokenizers-ggml-vocabs .......   Passed    0.47 sec
      Start 17: test-sampling
17/45 Test #17: test-sampling .....................   Passed    0.86 sec
      Start 18: test-reasoning-budget
18/45 Test #18: test-reasoning-budget .............   Passed    0.00 sec
      Start 19: test-grammar-parser
19/45 Test #19: test-grammar-parser ...............   Passed    0.00 sec
      Start 20: test-grammar-integration
20/45 Test #20: test-grammar-integration ..........   Passed    0.01 sec
      Start 21: test-llama-grammar
21/45 Test #21: test-llama-grammar ................   Passed    0.00 sec
      Start 22: test-chat
22/45 Test #22: test-chat .........................   Passed    2.11 sec
      Start 23: test-json-schema-to-grammar
23/45 Test #23: test-json-schema-to-grammar .......   Passed    1.91 sec
      Start 24: test-tokenizer-1-llama-spm
24/45 Test #24: test-tokenizer-1-llama-spm ........   Passed    0.11 sec
      Start 25: test-llama-archs
25/45 Test #25: test-llama-archs ..................   Passed    4.37 sec
      Start 26: test-chat-peg-parser
26/45 Test #26: test-chat-peg-parser ..............   Passed    0.02 sec
      Start 27: test-jinja
27/45 Test #27: test-jinja ........................   Passed    0.06 sec
      Start 28: test-jinja-py
28/45 Test #28: test-jinja-py .....................   Passed   11.41 sec
      Start 29: test-chat-auto-parser
29/45 Test #29: test-chat-auto-parser .............   Passed    0.28 sec
      Start 30: test-chat-template
30/45 Test #30: test-chat-template ................   Passed    0.06 sec
      Start 31: test-json-partial
31/45 Test #31: test-json-partial .................   Passed    0.00 sec
      Start 32: test-log
32/45 Test #32: test-log ..........................   Passed    0.01 sec
      Start 33: test-peg-parser
33/45 Test #33: test-peg-parser ...................   Passed    0.01 sec
      Start 34: test-regex-partial
34/45 Test #34: test-regex-partial ................   Passed    0.00 sec
      Start 35: test-download-model
35/45 Test #35: test-download-model ...............   Passed    2.15 sec
      Start 36: test-thread-safety
36/45 Test #36: test-thread-safety ................   Passed    0.25 sec
      Start 37: test-arg-parser
37/45 Test #37: test-arg-parser ...................   Passed    0.08 sec
      Start 38: test-gguf
38/45 Test #38: test-gguf .........................   Passed    0.06 sec
      Start 39: test-backend-ops
39/45 Test #39: test-backend-ops ..................   Passed    0.01 sec
      Start 46: test-barrier
40/45 Test #46: test-barrier ......................   Passed    0.27 sec
      Start 47: test-quantize-fns
41/45 Test #47: test-quantize-fns .................   Passed    1.05 sec
      Start 48: test-quantize-perf
42/45 Test #48: test-quantize-perf ................   Passed    0.05 sec
      Start 49: test-rope
43/45 Test #49: test-rope .........................   Passed    0.04 sec
      Start 50: test-mtmd-c-api
44/45 Test #50: test-mtmd-c-api ...................   Passed    0.00 sec
      Start 51: test-alloc
45/45 Test #51: test-alloc ........................   Passed    0.00 sec

100% tests passed, 0 tests failed out of 45

Label Time Summary:
main      =  14.34 sec*proc (43 tests)
python    =  11.41 sec*proc (1 test)

Total Test time (real) =  27.92 sec

real	0m27,932s
user	0m48,059s
sys	0m3,839s

embd_bge_small

BGE Small (BERT):

  • status: 0
  • f16:
+ ./bin/llama-embedding --model ../models-mnt/bge-small/ggml-model-f16.gguf -p 'I believe the meaning of life is' -ngl 99 -c 0 --no-op-offload
warning: no usable GPU found, --gpu-layers option will be ignored
warning: one possible reason is that llama.cpp was compiled without GPU support
warning: consult docs/build.md for compilation instructions
0.00.001.751 I main: n_parallel == 1 -> unified KV cache is enabled
0.00.001.767 I common_init_result: fitting params to device memory ...
0.00.001.768 I common_init_result: (for bugs during this step try to reproduce them with -fit off, or provide --verbose logs if the bug only occurs with -fit on)
0.00.037.794 I common_params_fit_impl: projected to use 69 MiB of host memory vs. 77898 MiB of total host memory
0.00.052.091 W load: special_eos_id is not in special_eog_ids - the tokenizer config may be incorrect
0.00.079.876 I common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
0.00.093.002 I 
0.00.093.087 I system_info: n_threads = 8 (n_threads_batch = 8) / 16 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 | 
0.00.094.049 I batch_decode: n_tokens = 9, n_seq = 1

embedding 0: -0.043960 -0.019931  0.007712 -0.000751  0.001344 -0.036999  0.109418  0.042582  0.092107 -0.015945  0.006784 -0.035670 -0.017907  0.015017  0.018081  0.015926 -0.011260  0.010490 -0.085274 -0.008459  0.091390 -0.017088 -0.060390 -0.024478  0.027519  0.076056  0.027986 -0.014574  0.017650 -0.033289 -0.037877 -0.019006  0.068682 -0.009854 -0.025034  0.072386 -0.046555  0.011033 -0.050286  0.047710  0.032437 -0.011726  0.022050  0.049604  0.010482  0.005781 -0.028890  0.008935 -0.018469 -0.051451 -0.046061  0.030487 -0.035441  0.054225 -0.069635  0.044224  0.029798  0.046302  0.073453 -0.042601  0.076094  0.038833 -0.181188  0.082507  0.042264 -0.064571 -0.060111 -0.017840  0.006475  0.005886  0.017158 -0.026637  0.064610  0.112595  0.035123 -0.067396  0.027069 -0.067289 -0.033462 -0.033188  0.033237  0.013539 -0.003328 -0.037494 -0.052071  0.055166 -0.002017 -0.038255  0.064464  0.028839 -0.043310 -0.029211 -0.039469  0.036347  0.008376 -0.015456 -0.036623  0.018123  0.028586  0.342812 -0.044411  0.056112  0.017662 -0.020885 -0.066804  0.000129 -0.037850 -0.030067 -0.008578 -0.021541  0.000520 -0.003230  0.004019  0.018876 -0.008559  0.025835  0.049417  0.000103  0.050908 -0.042491 -0.031902  0.023618  0.030714 -0.023199 -0.046255 -0.079230  0.115155  0.046709  0.027811 -0.040727  0.067800 -0.022945  0.010306 -0.032945 -0.018295  0.043837  0.024285  0.052386  0.007493  0.008896  0.011247 -0.074669 -0.065547 -0.026802 -0.041168 -0.023887  0.026716  0.006903  0.027724  0.052860 -0.036699  0.057693 -0.000230  0.031761 -0.019740 -0.022072  0.041010 -0.058929  0.019603  0.043151  0.043615  0.041618 -0.022525  0.027063 -0.021866  0.005424 -0.041325 -0.001270  0.024436  0.002108  0.044329 -0.022718  0.043624  0.064766  0.055420  0.037081 -0.000899  0.046123  0.045816 -0.008500  0.063064 -0.073204 -0.011986  0.032123  0.023920  0.014668 -0.033710  0.001100 -0.015855 -0.019007  0.047854  0.110846  0.028421  0.031361 -0.013281 -0.057457  0.006676  0.005131 -0.012261 -0.051530 -0.000929 -0.017667 -0.019422 -0.040945  0.009207 -0.057959  0.050955  0.052410 -0.009574 -0.040242 -0.014044 -0.024869 -0.017278  0.006302  0.006588 -0.026899  0.015624  0.030757  0.002582  0.023228 -0.022214 -0.098589 -0.051112 -0.278050 -0.014963 -0.061568 -0.027236  0.017703 -0.010968 -0.017083  0.035028  0.046984 -0.015453  0.015220 -0.025476  0.047851 -0.005930 -0.000693 -0.061024 -0.068902 -0.060359 -0.035918  0.043342 -0.055018  0.015045  0.000518 -0.058211 -0.010436  0.012567  0.151515  0.127119 -0.013597  0.041975 -0.025644  0.014008 -0.001054 -0.150439  0.044871  0.005309 -0.036256 -0.029792 -0.020171 -0.034870  0.010267  0.033580 -0.048145 -0.051801 -0.017462 -0.023506  0.047393  0.052029 -0.016825 -0.055483  0.025833 -0.005685  0.010733  0.038723  0.008216 -0.009770 -0.105720 -0.027463 -0.096157  0.025080 -0.011272  0.092367  0.056073  0.003696  0.027793  0.002038 -0.051117 -0.039891 -0.013549 -0.044969 -0.015316  0.002913 -0.043493 -0.077962  0.065198 -0.006816 -0.001589 -0.014637  0.071582  0.023727 -0.037145  0.009170  0.001555 -0.032286  0.015427  0.037899  0.000311 -0.053185  0.021337 -0.039844  0.000057  0.013402  0.019822 -0.057872  0.006451 -0.049526 -0.267880  0.039165 -0.067962  0.038210 -0.012318  0.041493 -0.016126  0.052432 -0.071385  0.011312  0.024689 -0.007221  0.082055  0.028516 -0.021516  0.040582 -0.004526 -0.074635 -0.014780  0.019998  0.002250  0.023125  0.197151 -0.043246 -0.025972 -0.004938 -0.019236  0.074234  0.001751 -0.031946 -0.036566 -0.045049  0.000551 -0.011591  0.018178 -0.029435 -0.008457  0.006441  0.050757 -0.014900  0.006181  0.026086 -0.030793  0.048078  0.114039 -0.040842 -0.011476  0.005396 -0.003585  0.025123 -0.059147  0.013750 -0.010379  0.038713  0.051474  0.035410  0.035039 -0.017055  0.026369 -0.014503 -0.050014  0.003251  0.054109  0.039720 -0.039147 


real	0m0,109s
user	0m0,179s
sys	0m0,034s
  • q8_0:
+ ./bin/llama-embedding --model ../models-mnt/bge-small/ggml-model-q8_0.gguf -p 'I believe the meaning of life is' -ngl 99 -c 0 --no-op-offload
warning: no usable GPU found, --gpu-layers option will be ignored
warning: one possible reason is that llama.cpp was compiled without GPU support
warning: consult docs/build.md for compilation instructions
0.00.001.629 I main: n_parallel == 1 -> unified KV cache is enabled
0.00.001.638 I common_init_result: fitting params to device memory ...
0.00.001.639 I common_init_result: (for bugs during this step try to reproduce them with -fit off, or provide --verbose logs if the bug only occurs with -fit on)
0.00.038.033 I common_params_fit_impl: projected to use 40 MiB of host memory vs. 77898 MiB of total host memory
0.00.052.143 W load: special_eos_id is not in special_eog_ids - the tokenizer config may be incorrect
0.00.077.834 I common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
0.00.087.543 I 
0.00.087.621 I system_info: n_threads = 8 (n_threads_batch = 8) / 16 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 | 
0.00.088.625 I batch_decode: n_tokens = 9, n_seq = 1

embedding 0: -0.044769 -0.020293  0.008312 -0.001913  0.002951 -0.036361  0.109152  0.042860  0.091707 -0.015266  0.006223 -0.036915 -0.018787  0.014956  0.017155  0.014664 -0.013389  0.010356 -0.085124 -0.007760  0.092818 -0.017684 -0.061927 -0.024778  0.027992  0.076285  0.027446 -0.014928  0.017682 -0.033781 -0.038384 -0.017640  0.068602 -0.010417 -0.023592  0.072539 -0.046256  0.010643 -0.051011  0.049985  0.032044 -0.012679  0.021985  0.050196  0.010734  0.005425 -0.027972  0.008380 -0.018950 -0.052749 -0.045274  0.028617 -0.035570  0.052332 -0.068761  0.043724  0.029066  0.045856  0.073537 -0.043214  0.075989  0.038740 -0.182764  0.082396  0.043152 -0.065902 -0.059464 -0.017057  0.006858  0.004254  0.016762 -0.027090  0.066127  0.112956  0.033962 -0.067310  0.026852 -0.066913 -0.034776 -0.034429  0.033462  0.014505 -0.003978 -0.037120 -0.051974  0.054514 -0.003063 -0.037339  0.063217  0.028557 -0.042307 -0.028936 -0.039699  0.036995  0.007952 -0.014911 -0.035936  0.018657  0.030676  0.344795 -0.044852  0.056983  0.017340 -0.021102 -0.063378 -0.000059 -0.037958 -0.029928 -0.008731 -0.020472  0.001127 -0.004184  0.004079  0.018009 -0.009781  0.024267  0.048652 -0.001033  0.051251 -0.041336 -0.030359  0.023049  0.030682 -0.023500 -0.045304 -0.078945  0.114234  0.048049  0.027377 -0.041017  0.067538 -0.022838  0.011250 -0.033847 -0.016738  0.044509  0.022050  0.052398  0.007432  0.006576  0.010436 -0.075216 -0.064472 -0.026449 -0.041538 -0.024660  0.026852  0.006176  0.026528  0.051708 -0.037361  0.058659  0.001662  0.032153 -0.019548 -0.021730  0.041598 -0.059513  0.019495  0.042868  0.042108  0.040890 -0.022224  0.028439 -0.022892  0.006937 -0.041591  0.000471  0.024398  0.001802  0.043274 -0.023550  0.043495  0.064700  0.056079  0.038828 -0.000302  0.047336  0.045703 -0.009004  0.060762 -0.072220 -0.011673  0.033569  0.023710  0.014491 -0.033240  0.001091 -0.016882 -0.017800  0.047999  0.110772  0.028647  0.030924 -0.010618 -0.056152  0.005752  0.004509 -0.011819 -0.051905 -0.002408 -0.017285 -0.020138 -0.040943  0.009639 -0.058830  0.051337  0.051765 -0.010950 -0.039967 -0.015824 -0.024824 -0.015123  0.005632  0.006604 -0.027472  0.015294  0.031295  0.002561  0.023642 -0.021440 -0.098875 -0.050686 -0.277322 -0.013596 -0.060241 -0.027209  0.017223 -0.008998 -0.017215  0.034745  0.048475 -0.016189  0.015411 -0.023170  0.049729 -0.005031  0.000517 -0.060754 -0.069485 -0.060687 -0.035264  0.045340 -0.056031  0.014686 -0.000107 -0.059174 -0.010550  0.010932  0.150441  0.126988 -0.012928  0.043261 -0.026197  0.014554 -0.001012 -0.148888  0.043045  0.005783 -0.035040 -0.029486 -0.019778 -0.034115  0.009486  0.035182 -0.048500 -0.052967 -0.016393 -0.024666  0.048341  0.050784 -0.017477 -0.056926  0.023685 -0.005170  0.011758  0.038203  0.006412 -0.008393 -0.105756 -0.028217 -0.097797  0.024940 -0.011161  0.091588  0.056131  0.004987  0.027713  0.001477 -0.050231 -0.038585 -0.013572 -0.046539 -0.015327  0.002897 -0.043732 -0.077053  0.065700 -0.006396 -0.000981 -0.015114  0.070910  0.024637 -0.036404  0.008641  0.001068 -0.033026  0.016985  0.037254  0.000248 -0.052019  0.020569 -0.038401 -0.000725  0.012999  0.019775 -0.058274  0.004633 -0.049796 -0.268352  0.039090 -0.067125  0.036869 -0.011389  0.043082 -0.016243  0.050856 -0.071612  0.012399  0.024082 -0.007879  0.083606  0.029305 -0.021130  0.042402 -0.003813 -0.073862 -0.014481  0.019215  0.001791  0.023348  0.196663 -0.044314 -0.026124 -0.004931 -0.019014  0.074743  0.001652 -0.032484 -0.037155 -0.045355  0.001014 -0.011051  0.017752 -0.026512 -0.009489  0.005867  0.050566 -0.014158  0.005610  0.027206 -0.031022  0.048065  0.112428 -0.040627 -0.011999  0.004289 -0.002847  0.025315 -0.060347  0.014617 -0.010438  0.037544  0.049858  0.036079  0.036533 -0.017197  0.025771 -0.014269 -0.049203  0.003926  0.053891  0.040036 -0.038903 


real	0m0,102s
user	0m0,145s
sys	0m0,036s

rerank_tiny

Rerank Tiny (Jina):

  • status: 0
  • f16:
+ ./bin/llama-embedding --model ../models-mnt/rerank-tiny/ggml-model-f16.gguf -p 'what is panda?\thi\nwhat is panda?\tit'\''s a bear\nwhat is panda?\tThe giant panda (Ailuropoda melanoleuca), sometimes called a panda bear or simply panda, is a bear species endemic to China.' -ngl 99 -c 0 --pooling rank --embd-normalize -1 --no-op-offload --verbose-prompt
warning: no usable GPU found, --gpu-layers option will be ignored
warning: one possible reason is that llama.cpp was compiled without GPU support
warning: consult docs/build.md for compilation instructions
0.00.002.494 I main: n_parallel == 1 -> unified KV cache is enabled
0.00.002.505 I common_init_result: fitting params to device memory ...
0.00.002.505 I common_init_result: (for bugs during this step try to reproduce them with -fit off, or provide --verbose logs if the bug only occurs with -fit on)
0.00.271.161 I common_params_fit_impl: projected to use 115 MiB of host memory vs. 77898 MiB of total host memory
0.00.286.548 W load: empty token at index 5
0.00.290.813 W load: model vocab missing newline token, using special_pad_id instead
0.00.508.456 W llama_init_from_model: model default pooling_type is [-1], but [4] was specified
0.00.531.030 I common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
0.00.540.564 I 
0.00.540.650 I system_info: n_threads = 8 (n_threads_batch = 8) / 16 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 | 
0.00.540.829 I main: prompt 0: 'what is panda?	hi'
0.00.540.831 I main: number of tokens in prompt = 9
     0 -> '<s>'
 21381 -> 'what'
 21152 -> 'is'
 49990 -> 'panda'
    61 -> '?'
     2 -> '</s>'
     2 -> '</s>'
 23233 -> 'hi'
     2 -> '</s>'


0.00.540.838 I main: prompt 1: 'what is panda?	it's a bear'
0.00.540.838 I main: number of tokens in prompt = 13
     0 -> '<s>'
 21381 -> 'what'
 21152 -> 'is'
 49990 -> 'panda'
    61 -> '?'
     2 -> '</s>'
     2 -> '</s>'
 21153 -> 'it'
    37 -> '''
    87 -> 's'
    69 -> 'a'
 25706 -> 'bear'
     2 -> '</s>'


0.00.540.840 I main: prompt 2: 'what is panda?	The giant panda (Ailuropoda melanoleuca), sometimes called a panda bear or simply panda, is a bear species endemic to China.'
0.00.540.841 I main: number of tokens in prompt = 40
     0 -> '<s>'
 21381 -> 'what'
 21152 -> 'is'
 49990 -> 'panda'
    61 -> '?'
     2 -> '</s>'
     2 -> '</s>'
 21215 -> 'he'
 28390 -> 'giant'
 49990 -> 'panda'
    38 -> '('
 21163 -> 'il'
 26237 -> 'uro'
 21223 -> 'po'
 23179 -> 'da'
 36906 -> 'melan'
 26791 -> 'ole'
    89 -> 'u'
 21402 -> 'ca'
 21686 -> '),'
 23314 -> 'sometimes'
 22517 -> 'called'
    69 -> 'a'
 49990 -> 'panda'
 25706 -> 'bear'
 21142 -> 'or'
 22810 -> 'simply'
 49990 -> 'panda'
    42 -> ','
 21152 -> 'is'
    69 -> 'a'
 25706 -> 'bear'
 25677 -> 'species'
 28930 -> 'ende'
 22024 -> 'mic'
 21148 -> 'to'
    76 -> 'h'
 22344 -> 'ina'
    44 -> '.'
     2 -> '</s>'


0.00.540.912 I batch_decode: n_tokens = 62, n_seq = 3

rerank score 0:    0.023
rerank score 1:    0.025
rerank score 2:    0.199


real	0m0,563s
user	0m0,579s
sys	0m0,078s
  - rerank score 0 @ 0.023 OK
  - rerank score 1 @ 0.025 OK
  - rerank score 2 @ 0.199 OK

test_scripts

Runs test scripts

  • status: 0
+ cd ./tools/gguf-split
+ bash tests.sh /root/llama.me/build-ci-release/bin /root/llama.me/tmp/mnt/models
+ SPLIT=/root/llama.me/build-ci-release/bin/llama-gguf-split
+ MAIN=/root/llama.me/build-ci-release/bin/llama-completion
+ WORK_PATH=/root/llama.me/tmp/mnt/models/gguf-split
+++ dirname tests.sh
++ realpath ./../../
+ ROOT_DIR=/root/llama.me
+ mkdir -p /root/llama.me/tmp/mnt/models/gguf-split
+ rm -f '/root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split*.gguf' '/root/llama.me/tmp/mnt/models/gguf-split/ggml-model-merge*.gguf'
+ cd /root/llama.me/tmp/mnt/models/gguf-split
+ /root/llama.me/scripts/hf.sh --repo ggml-org/Qwen3-0.6B-GGUF --file Qwen3-0.6B-Q8_0.gguf
[+] attempting to download Qwen3-0.6B-Q8_0.gguf
[+] wget -q -c -O ./Qwen3-0.6B-Q8_0.gguf https://huggingface.co/ggml-org/Qwen3-0.6B-GGUF/resolve/main/Qwen3-0.6B-Q8_0.gguf
./Qwen3-0.6B-Q8_0.gguf
+ echo PASS
PASS
+ /root/llama.me/build-ci-release/bin/llama-gguf-split --split-max-tensors 28 /root/llama.me/tmp/mnt/models/gguf-split/Qwen3-0.6B-Q8_0.gguf /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split
n_split: 12
split 00001: n_tensors = 28, total_size = 371M
split 00002: n_tensors = 28, total_size = 45M
split 00003: n_tensors = 28, total_size = 40M
split 00004: n_tensors = 28, total_size = 43M
split 00005: n_tensors = 28, total_size = 42M
split 00006: n_tensors = 28, total_size = 44M
split 00007: n_tensors = 28, total_size = 39M
split 00008: n_tensors = 28, total_size = 45M
split 00009: n_tensors = 28, total_size = 39M
split 00010: n_tensors = 28, total_size = 44M
split 00011: n_tensors = 28, total_size = 42M
split 00012: n_tensors = 3, total_size = 6M
Writing file /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-00001-of-00012.gguf ... done
Writing file /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-00002-of-00012.gguf ... done
Writing file /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-00003-of-00012.gguf ... done
Writing file /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-00004-of-00012.gguf ... done
Writing file /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-00005-of-00012.gguf ... done
Writing file /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-00006-of-00012.gguf ... done
Writing file /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-00007-of-00012.gguf ... done
Writing file /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-00008-of-00012.gguf ... done
Writing file /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-00009-of-00012.gguf ... done
Writing file /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-00010-of-00012.gguf ... done
Writing file /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-00011-of-00012.gguf ... done
Writing file /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-00012-of-00012.gguf ... gguf_split: 12 gguf split written with a total of 311 tensors.
done
+ echo PASS
PASS
+ echo

+ /root/llama.me/build-ci-release/bin/llama-completion -no-cnv --model /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-00001-of-00012.gguf -p 'I believe the meaning of life is' --n-predict 32
0.00.001.768 I llama_completion: llama backend init
0.00.001.773 I llama_completion: load the model and apply lora adapter, if any
0.00.001.779 I common_init_result: fitting params to device memory ...
0.00.001.780 I common_init_result: (for bugs during this step try to reproduce them with -fit off, or provide --verbose logs if the bug only occurs with -fit on)
0.00.184.861 I common_params_fit_impl: projected to use 5542 MiB of host memory vs. 77898 MiB of total host memory
0.00.283.379 W load: control-looking token: 128247 '</s>' was not control-type; this is probably a bug in the model. its type will be overridden
0.01.968.347 I common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
0.02.047.906 I llama_completion: llama threadpool init, n_threads = 8
0.02.047.924 I 
0.02.047.984 I system_info: n_threads = 8 (n_threads_batch = 8) / 16 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 | 
0.02.047.985 I 
0.02.048.054 I sampler seed: 1465155874
0.02.048.062 I sampler params: 
	repeat_last_n = 64, repeat_penalty = 1,000, frequency_penalty = 0,000, presence_penalty = 0,000
	dry_multiplier = 0,000, dry_base = 1,750, dry_allowed_length = 2, dry_penalty_last_n = -1
	top_k = 40, top_p = 0,950, min_p = 0,050, xtc_probability = 0,000, xtc_threshold = 0,100, typical_p = 1,000, top_n_sigma = -1,000, temp = 0,800
	mirostat = 0, mirostat_lr = 0,100, mirostat_ent = 5,000, adaptive_target = -1,000, adaptive_decay = 0,900
0.02.048.067 I sampler chain: logits -> ?penalties -> ?dry -> ?top-n-sigma -> top-k -> ?typical -> top-p -> min-p -> ?xtc -> temp-ext -> dist 
0.02.048.068 I generate: n_ctx = 40960, n_batch = 2048, n_predict = 32, n_keep = 0
0.02.048.068 I 
I believe the meaning of life is to find one's true self, and I can't see myself as such. I feel this is a problem of the mind, which is separate from the external

0.02.453.878 I common_perf_print:    sampling time =       4,82 ms
0.02.453.879 I common_perf_print:    samplers time =       1,68 ms /    39 tokens
0.02.453.883 I common_perf_print:        load time =    1824,32 ms
0.02.453.884 I common_perf_print: prompt eval time =      20,72 ms /     7 tokens (    2,96 ms per token,   337,81 tokens per second)
0.02.453.885 I common_perf_print:        eval time =     379,52 ms /    31 runs   (   12,24 ms per token,    81,68 tokens per second)
0.02.453.885 I common_perf_print:       total time =     405,98 ms /    38 tokens
0.02.453.885 I common_perf_print: unaccounted time =       0,91 ms /   0,2 %      (total - sampling - prompt eval - eval) / (total)
0.02.453.886 I common_perf_print:    graphs reused =         30
+ echo PASS
PASS
+ echo

+ /root/llama.me/build-ci-release/bin/llama-gguf-split --merge /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-00001-of-00012.gguf /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-merge.gguf
gguf_merge: /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-00001-of-00012.gguf -> /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-merge.gguf
gguf_merge: reading metadata /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-00001-of-00012.gguf ...�[3Ddone
gguf_merge: reading metadata /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-00002-of-00012.gguf ...�[3Ddone
gguf_merge: reading metadata /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-00003-of-00012.gguf ...�[3Ddone
gguf_merge: reading metadata /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-00004-of-00012.gguf ...�[3Ddone
gguf_merge: reading metadata /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-00005-of-00012.gguf ...�[3Ddone
gguf_merge: reading metadata /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-00006-of-00012.gguf ...�[3Ddone
gguf_merge: reading metadata /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-00007-of-00012.gguf ...�[3Ddone
gguf_merge: reading metadata /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-00008-of-00012.gguf ...�[3Ddone
gguf_merge: reading metadata /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-00009-of-00012.gguf ...�[3Ddone
gguf_merge: reading metadata /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-00010-of-00012.gguf ...�[3Ddone
gguf_merge: reading metadata /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-00011-of-00012.gguf ...�[3Ddone
gguf_merge: reading metadata /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-00012-of-00012.gguf ...�[3Ddone
gguf_merge: writing tensors /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-00001-of-00012.gguf ...�[3Ddone
gguf_merge: writing tensors /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-00002-of-00012.gguf ...�[3Ddone
gguf_merge: writing tensors /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-00003-of-00012.gguf ...�[3Ddone
gguf_merge: writing tensors /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-00004-of-00012.gguf ...�[3Ddone
gguf_merge: writing tensors /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-00005-of-00012.gguf ...�[3Ddone
gguf_merge: writing tensors /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-00006-of-00012.gguf ...�[3Ddone
gguf_merge: writing tensors /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-00007-of-00012.gguf ...�[3Ddone
gguf_merge: writing tensors /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-00008-of-00012.gguf ...�[3Ddone
gguf_merge: writing tensors /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-00009-of-00012.gguf ...�[3Ddone
gguf_merge: writing tensors /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-00010-of-00012.gguf ...�[3Ddone
gguf_merge: writing tensors /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-00011-of-00012.gguf ...�[3Ddone
gguf_merge: writing tensors /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-00012-of-00012.gguf ...�[3Ddone
gguf_merge: /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-merge.gguf merged from 12 split with 311 tensors.
+ echo PASS
PASS
+ echo

+ /root/llama.me/build-ci-release/bin/llama-completion -no-cnv --model /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-merge.gguf -p 'I believe the meaning of life is' --n-predict 32
0.00.001.773 I llama_completion: llama backend init
0.00.001.777 I llama_completion: load the model and apply lora adapter, if any
0.00.001.784 I common_init_result: fitting params to device memory ...
0.00.001.785 I common_init_result: (for bugs during this step try to reproduce them with -fit off, or provide --verbose logs if the bug only occurs with -fit on)
0.00.196.093 I common_params_fit_impl: projected to use 5542 MiB of host memory vs. 77898 MiB of total host memory
0.00.279.046 W load: control-looking token: 128247 '</s>' was not control-type; this is probably a bug in the model. its type will be overridden
0.01.976.584 I common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
0.02.057.590 I llama_completion: llama threadpool init, n_threads = 8
0.02.057.609 I 
0.02.057.665 I system_info: n_threads = 8 (n_threads_batch = 8) / 16 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 | 
0.02.057.666 I 
0.02.057.733 I sampler seed: 774352655
0.02.057.740 I sampler params: 
	repeat_last_n = 64, repeat_penalty = 1,000, frequency_penalty = 0,000, presence_penalty = 0,000
	dry_multiplier = 0,000, dry_base = 1,750, dry_allowed_length = 2, dry_penalty_last_n = -1
	top_k = 40, top_p = 0,950, min_p = 0,050, xtc_probability = 0,000, xtc_threshold = 0,100, typical_p = 1,000, top_n_sigma = -1,000, temp = 0,800
	mirostat = 0, mirostat_lr = 0,100, mirostat_ent = 5,000, adaptive_target = -1,000, adaptive_decay = 0,900
0.02.057.749 I sampler chain: logits -> ?penalties -> ?dry -> ?top-n-sigma -> top-k -> ?typical -> top-p -> min-p -> ?xtc -> temp-ext -> dist 
0.02.057.750 I generate: n_ctx = 40960, n_batch = 2048, n_predict = 32, n_keep = 0
0.02.057.750 I 
I believe the meaning of life is to pursue what is important and achieve what is necessary. That is to say, we should always look for the meaning we find and pursue it. However, the

0.02.462.880 I common_perf_print:    sampling time =       4,67 ms
0.02.462.881 I common_perf_print:    samplers time =       1,62 ms /    39 tokens
0.02.462.885 I common_perf_print:        load time =    1822,33 ms
0.02.462.886 I common_perf_print: prompt eval time =      16,54 ms /     7 tokens (    2,36 ms per token,   423,11 tokens per second)
0.02.462.887 I common_perf_print:        eval time =     383,09 ms /    31 runs   (   12,36 ms per token,    80,92 tokens per second)
0.02.462.887 I common_perf_print:       total time =     405,30 ms /    38 tokens
0.02.462.888 I common_perf_print: unaccounted time =       1,00 ms /   0,2 %      (total - sampling - prompt eval - eval) / (total)
0.02.462.888 I common_perf_print:    graphs reused =         30
+ echo PASS
PASS
+ echo

+ /root/llama.me/build-ci-release/bin/llama-gguf-split --split-max-tensors 32 --no-tensor-first-split /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-merge.gguf /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-32-tensors
n_split: 11
split 00001: n_tensors = 0, total_size = 5M
split 00002: n_tensors = 32, total_size = 370M
split 00003: n_tensors = 32, total_size = 49M
split 00004: n_tensors = 32, total_size = 50M
split 00005: n_tensors = 32, total_size = 47M
split 00006: n_tensors = 32, total_size = 47M
split 00007: n_tensors = 32, total_size = 50M
split 00008: n_tensors = 32, total_size = 50M
split 00009: n_tensors = 32, total_size = 49M
split 00010: n_tensors = 32, total_size = 46M
split 00011: n_tensors = 23, total_size = 36M
Writing file /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-32-tensors-00001-of-00011.gguf ... done
Writing file /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-32-tensors-00002-of-00011.gguf ... done
Writing file /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-32-tensors-00003-of-00011.gguf ... done
Writing file /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-32-tensors-00004-of-00011.gguf ... done
Writing file /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-32-tensors-00005-of-00011.gguf ... done
Writing file /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-32-tensors-00006-of-00011.gguf ... done
Writing file /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-32-tensors-00007-of-00011.gguf ... done
Writing file /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-32-tensors-00008-of-00011.gguf ... done
Writing file /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-32-tensors-00009-of-00011.gguf ... done
Writing file /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-32-tensors-00010-of-00011.gguf ... done
Writing file /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-32-tensors-00011-of-00011.gguf ... gguf_split: 11 gguf split written with a total of 311 tensors.
done
+ echo PASS
PASS
+ echo

+ /root/llama.me/build-ci-release/bin/llama-completion -no-cnv --model /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-32-tensors-00001-of-00011.gguf -p 'I believe the meaning of life is' --n-predict 32
0.00.001.769 I llama_completion: llama backend init
0.00.001.774 I llama_completion: load the model and apply lora adapter, if any
0.00.001.781 I common_init_result: fitting params to device memory ...
0.00.001.782 I common_init_result: (for bugs during this step try to reproduce them with -fit off, or provide --verbose logs if the bug only occurs with -fit on)
0.00.175.397 I common_params_fit_impl: projected to use 5542 MiB of host memory vs. 77898 MiB of total host memory
0.00.255.197 W load: control-looking token: 128247 '</s>' was not control-type; this is probably a bug in the model. its type will be overridden
0.01.939.205 I common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
0.02.018.293 I llama_completion: llama threadpool init, n_threads = 8
0.02.018.310 I 
0.02.018.370 I system_info: n_threads = 8 (n_threads_batch = 8) / 16 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 | 
0.02.018.371 I 
0.02.018.451 I sampler seed: 1386843486
0.02.018.458 I sampler params: 
	repeat_last_n = 64, repeat_penalty = 1,000, frequency_penalty = 0,000, presence_penalty = 0,000
	dry_multiplier = 0,000, dry_base = 1,750, dry_allowed_length = 2, dry_penalty_last_n = -1
	top_k = 40, top_p = 0,950, min_p = 0,050, xtc_probability = 0,000, xtc_threshold = 0,100, typical_p = 1,000, top_n_sigma = -1,000, temp = 0,800
	mirostat = 0, mirostat_lr = 0,100, mirostat_ent = 5,000, adaptive_target = -1,000, adaptive_decay = 0,900
0.02.018.463 I sampler chain: logits -> ?penalties -> ?dry -> ?top-n-sigma -> top-k -> ?typical -> top-p -> min-p -> ?xtc -> temp-ext -> dist 
0.02.018.463 I generate: n_ctx = 40960, n_batch = 2048, n_predict = 32, n_keep = 0
0.02.018.463 I 
I believe the meaning of life is to be happy, and I believe that happiness is a result of a great deal of personal effort to be a happy person. This belief is supported by the fact

0.02.427.636 I common_perf_print:    sampling time =       5,22 ms
0.02.427.638 I common_perf_print:    samplers time =       1,83 ms /    39 tokens
0.02.427.643 I common_perf_print:        load time =    1804,39 ms
0.02.427.644 I common_perf_print: prompt eval time =      20,66 ms /     7 tokens (    2,95 ms per token,   338,75 tokens per second)
0.02.427.645 I common_perf_print:        eval time =     382,49 ms /    31 runs   (   12,34 ms per token,    81,05 tokens per second)
0.02.427.645 I common_perf_print:       total time =     409,35 ms /    38 tokens
0.02.427.646 I common_perf_print: unaccounted time =       0,98 ms /   0,2 %      (total - sampling - prompt eval - eval) / (total)
0.02.427.646 I common_perf_print:    graphs reused =         30
+ echo PASS
PASS
+ echo

+ /root/llama.me/build-ci-release/bin/llama-gguf-split --split-max-size 500M /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-merge.gguf /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-500M
n_split: 2
split 00001: n_tensors = 116, total_size = 504M
split 00002: n_tensors = 195, total_size = 299M
Writing file /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-500M-00001-of-00002.gguf ... done
Writing file /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-500M-00002-of-00002.gguf ... gguf_split: 2 gguf split written with a total of 311 tensors.
done
+ echo PASS
PASS
+ echo

+ /root/llama.me/build-ci-release/bin/llama-completion -no-cnv --model /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-500M-00001-of-00002.gguf -p 'I believe the meaning of life is' --n-predict 32
0.00.001.795 I llama_completion: llama backend init
0.00.001.799 I llama_completion: load the model and apply lora adapter, if any
0.00.001.805 I common_init_result: fitting params to device memory ...
0.00.001.805 I common_init_result: (for bugs during this step try to reproduce them with -fit off, or provide --verbose logs if the bug only occurs with -fit on)
0.00.198.644 I common_params_fit_impl: projected to use 5542 MiB of host memory vs. 77898 MiB of total host memory
0.00.283.923 W load: control-looking token: 128247 '</s>' was not control-type; this is probably a bug in the model. its type will be overridden
0.01.957.175 I common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
0.02.034.123 I llama_completion: llama threadpool init, n_threads = 8
0.02.034.147 I 
0.02.034.202 I system_info: n_threads = 8 (n_threads_batch = 8) / 16 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 | 
0.02.034.202 I 
0.02.034.268 I sampler seed: 400587886
0.02.034.278 I sampler params: 
	repeat_last_n = 64, repeat_penalty = 1,000, frequency_penalty = 0,000, presence_penalty = 0,000
	dry_multiplier = 0,000, dry_base = 1,750, dry_allowed_length = 2, dry_penalty_last_n = -1
	top_k = 40, top_p = 0,950, min_p = 0,050, xtc_probability = 0,000, xtc_threshold = 0,100, typical_p = 1,000, top_n_sigma = -1,000, temp = 0,800
	mirostat = 0, mirostat_lr = 0,100, mirostat_ent = 5,000, adaptive_target = -1,000, adaptive_decay = 0,900
0.02.034.283 I sampler chain: logits -> ?penalties -> ?dry -> ?top-n-sigma -> top-k -> ?typical -> top-p -> min-p -> ?xtc -> temp-ext -> dist 
0.02.034.283 I generate: n_ctx = 40960, n_batch = 2048, n_predict = 32, n_keep = 0
0.02.034.283 I 
I believe the meaning of life is to be happy, and that it is the way to the spiritual life. I want to find out if there is a way to achieve this. How should I

0.02.446.774 I common_perf_print:    sampling time =       5,32 ms
0.02.446.775 I common_perf_print:    samplers time =       1,68 ms /    39 tokens
0.02.446.779 I common_perf_print:        load time =    1797,11 ms
0.02.446.780 I common_perf_print: prompt eval time =      20,54 ms /     7 tokens (    2,93 ms per token,   340,82 tokens per second)
0.02.446.781 I common_perf_print:        eval time =     385,83 ms /    31 runs   (   12,45 ms per token,    80,35 tokens per second)
0.02.446.794 I common_perf_print:       total time =     412,65 ms /    38 tokens
0.02.446.795 I common_perf_print: unaccounted time =       0,96 ms /   0,2 %      (total - sampling - prompt eval - eval) / (total)
0.02.446.795 I common_perf_print:    graphs reused =         30
+ echo PASS
PASS
+ echo

+ rm -f /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-00001-of-00012.gguf /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-00002-of-00012.gguf /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-00003-of-00012.gguf /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-00004-of-00012.gguf /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-00005-of-00012.gguf /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-00006-of-00012.gguf /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-00007-of-00012.gguf /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-00008-of-00012.gguf /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-00009-of-00012.gguf /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-00010-of-00012.gguf /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-00011-of-00012.gguf /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-00012-of-00012.gguf /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-32-tensors-00001-of-00011.gguf /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-32-tensors-00002-of-00011.gguf /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-32-tensors-00003-of-00011.gguf /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-32-tensors-00004-of-00011.gguf /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-32-tensors-00005-of-00011.gguf /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-32-tensors-00006-of-00011.gguf /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-32-tensors-00007-of-00011.gguf /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-32-tensors-00008-of-00011.gguf /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-32-tensors-00009-of-00011.gguf /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-32-tensors-00010-of-00011.gguf /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-32-tensors-00011-of-00011.gguf /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-500M-00001-of-00002.gguf /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-split-500M-00002-of-00002.gguf /root/llama.me/tmp/mnt/models/gguf-split/ggml-model-merge.gguf

real	0m14,437s
user	0m16,790s
sys	0m9,612s
+ cd ./tools/quantize
+ bash tests.sh /root/llama.me/build-ci-release/bin /root/llama.me/tmp/mnt/models
+ SPLIT=/root/llama.me/build-ci-release/bin/llama-gguf-split
+ QUANTIZE=/root/llama.me/build-ci-release/bin/llama-quantize
+ MAIN=/root/llama.me/build-ci-release/bin/llama-completion
+ WORK_PATH=/root/llama.me/tmp/mnt/models/quantize
+++ dirname tests.sh
++ realpath ./../../
+ ROOT_DIR=/root/llama.me
+ mkdir -p /root/llama.me/tmp/mnt/models/quantize
+ rm -f '/root/llama.me/tmp/mnt/models/quantize/ggml-model-split*.gguf' '/root/llama.me/tmp/mnt/models/quantize/ggml-model-requant*.gguf'
+ cd /root/llama.me/tmp/mnt/models/quantize
+ /root/llama.me/scripts/hf.sh --repo ggml-org/Qwen3-0.6B-GGUF --file Qwen3-0.6B-Q8_0.gguf
[+] attempting to download Qwen3-0.6B-Q8_0.gguf
[+] wget -q -c -O ./Qwen3-0.6B-Q8_0.gguf https://huggingface.co/ggml-org/Qwen3-0.6B-GGUF/resolve/main/Qwen3-0.6B-Q8_0.gguf
./Qwen3-0.6B-Q8_0.gguf
+ echo PASS
PASS
+ /root/llama.me/build-ci-release/bin/llama-gguf-split --split-max-tensors 28 /root/llama.me/tmp/mnt/models/quantize/Qwen3-0.6B-Q8_0.gguf /root/llama.me/tmp/mnt/models/quantize/ggml-model-split
n_split: 12
split 00001: n_tensors = 28, total_size = 371M
split 00002: n_tensors = 28, total_size = 45M
split 00003: n_tensors = 28, total_size = 40M
split 00004: n_tensors = 28, total_size = 43M
split 00005: n_tensors = 28, total_size = 42M
split 00006: n_tensors = 28, total_size = 44M
split 00007: n_tensors = 28, total_size = 39M
split 00008: n_tensors = 28, total_size = 45M
split 00009: n_tensors = 28, total_size = 39M
split 00010: n_tensors = 28, total_size = 44M
split 00011: n_tensors = 28, total_size = 42M
split 00012: n_tensors = 3, total_size = 6M
Writing file /root/llama.me/tmp/mnt/models/quantize/ggml-model-split-00001-of-00012.gguf ... done
Writing file /root/llama.me/tmp/mnt/models/quantize/ggml-model-split-00002-of-00012.gguf ... done
Writing file /root/llama.me/tmp/mnt/models/quantize/ggml-model-split-00003-of-00012.gguf ... done
Writing file /root/llama.me/tmp/mnt/models/quantize/ggml-model-split-00004-of-00012.gguf ... done
Writing file /root/llama.me/tmp/mnt/models/quantize/ggml-model-split-00005-of-00012.gguf ... done
Writing file /root/llama.me/tmp/mnt/models/quantize/ggml-model-split-00006-of-00012.gguf ... done
Writing file /root/llama.me/tmp/mnt/models/quantize/ggml-model-split-00007-of-00012.gguf ... done
Writing file /root/llama.me/tmp/mnt/models/quantize/ggml-model-split-00008-of-00012.gguf ... done
Writing file /root/llama.me/tmp/mnt/models/quantize/ggml-model-split-00009-of-00012.gguf ... done
Writing file /root/llama.me/tmp/mnt/models/quantize/ggml-model-split-00010-of-00012.gguf ... done
Writing file /root/llama.me/tmp/mnt/models/quantize/ggml-model-split-00011-of-00012.gguf ... done
Writing file /root/llama.me/tmp/mnt/models/quantize/ggml-model-split-00012-of-00012.gguf ... gguf_split: 12 gguf split written with a total of 311 tensors.
done
+ echo PASS
PASS
+ echo

+ /root/llama.me/build-ci-release/bin/llama-quantize --allow-requantize --keep-split /root/llama.me/tmp/mnt/models/quantize/ggml-model-split-00001-of-00012.gguf /root/llama.me/tmp/mnt/models/quantize/ggml-model-requant.gguf Q4_K
llama_print_build_info: build = 9440 (eb91ed029)
llama_print_build_info: built with GNU 13.3.0 for Linux x86_64
llama_quantize: quantizing '/root/llama.me/tmp/mnt/models/quantize/ggml-model-split-00001-of-00012.gguf' to '/root/llama.me/tmp/mnt/models/quantize/ggml-model-requant' as Q4_K
llama_model_loader: additional 11 GGUFs metadata loaded.
llama_model_loader: loaded meta data with 37 key-value pairs and 311 tensors from /root/llama.me/tmp/mnt/models/quantize/ggml-model-split-00001-of-00012.gguf (version GGUF V3 (latest))
llama_model_loader: Dumping metadata keys/values. Note: KV overrides do not apply in this output.
llama_model_loader: - kv   0:                       general.architecture str              = qwen3
llama_model_loader: - kv   1:                               general.type str              = model
llama_model_loader: - kv   2:                               general.name str              = Qwen3 0.6B
llama_model_loader: - kv   3:                           general.basename str              = Qwen3
llama_model_loader: - kv   4:                         general.size_label str              = 0.6B
llama_model_loader: - kv   5:                            general.license str              = apache-2.0
llama_model_loader: - kv   6:                       general.license.link str              = https://huggingface.co/Qwen/Qwen3-0.6...
llama_model_loader: - kv   7:                   general.base_model.count u32              = 1
llama_model_loader: - kv   8:                  general.base_model.0.name str              = Qwen3 0.6B Base
llama_model_loader: - kv   9:          general.base_model.0.organization str              = Qwen
llama_model_loader: - kv  10:              general.base_model.0.repo_url str              = https://huggingface.co/Qwen/Qwen3-0.6...
llama_model_loader: - kv  11:                               general.tags arr[str,1]       = ["text-generation"]
llama_model_loader: - kv  12:                          qwen3.block_count u32              = 28
llama_model_loader: - kv  13:                       qwen3.context_length u32              = 40960
llama_model_loader: - kv  14:                     qwen3.embedding_length u32              = 1024
llama_model_loader: - kv  15:                  qwen3.feed_forward_length u32              = 3072
llama_model_loader: - kv  16:                 qwen3.attention.head_count u32              = 16
llama_model_loader: - kv  17:              qwen3.attention.head_count_kv u32              = 8
llama_model_loader: - kv  18:                       qwen3.rope.freq_base f32              = 1000000.000000
llama_model_loader: - kv  19:     qwen3.attention.layer_norm_rms_epsilon f32              = 0.000001
llama_model_loader: - kv  20:                 qwen3.attention.key_length u32              = 128
llama_model_loader: - kv  21:               qwen3.attention.value_length u32              = 128
llama_model_loader: - kv  22:                       tokenizer.ggml.model str              = gpt2
llama_model_loader: - kv  23:                         tokenizer.ggml.pre str              = qwen2
llama_model_loader: - kv  24:                      tokenizer.ggml.tokens arr[str,151936]  = ["!", "\"", "#", "$", "%", "&", "'", ...
llama_model_loader: - kv  25:                  tokenizer.ggml.token_type arr[i32,151936]  = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, ...
llama_model_loader: - kv  26:                      tokenizer.ggml.merges arr[str,151387]  = ["Ġ Ġ", "ĠĠ ĠĠ", "i n", "Ġ t",...
llama_model_loader: - kv  27:                tokenizer.ggml.eos_token_id u32              = 151645
llama_model_loader: - kv  28:            tokenizer.ggml.padding_token_id u32              = 151643
llama_model_loader: - kv  29:                tokenizer.ggml.bos_token_id u32              = 151643
llama_model_loader: - kv  30:               tokenizer.ggml.add_bos_token bool             = false
llama_model_loader: - kv  31:                    tokenizer.chat_template str              = {%- if tools %}\n    {{- '<|im_start|>...
llama_model_loader: - kv  32:               general.quantization_version u32              = 2
llama_model_loader: - kv  33:                          general.file_type u32              = 7
llama_model_loader: - kv  34:                                   split.no u16              = 0
llama_model_loader: - kv  35:                        split.tensors.count i32              = 311
llama_model_loader: - kv  36:                                split.count u16              = 12
llama_model_loader: - type  f32:  113 tensors
llama_model_loader: - type q8_0:  198 tensors
[   1/ 311] output.weight                        - [  1024, 151936,      1,      1], type =   q8_0, converting to q6_K .. size =   157.65 MiB ->   121.71 MiB
[   2/ 311] output_norm.weight                   - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[   3/ 311] token_embd.weight                    - [  1024, 151936,      1,      1], type =   q8_0, converting to q4_K .. size =   157.65 MiB ->    83.46 MiB
[   4/ 311] blk.0.attn_k.weight                  - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[   5/ 311] blk.0.attn_k_norm.weight             - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[   6/ 311] blk.0.attn_norm.weight               - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[   7/ 311] blk.0.attn_output.weight             - [  2048,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[   8/ 311] blk.0.attn_q.weight                  - [  1024,   2048,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[   9/ 311] blk.0.attn_q_norm.weight             - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[  10/ 311] blk.0.attn_v.weight                  - [  1024,   1024,      1,      1], type =   q8_0, converting to q6_K .. size =     1.06 MiB ->     0.82 MiB
[  11/ 311] blk.0.ffn_down.weight                - [  3072,   1024,      1,      1], type =   q8_0, converting to q6_K .. size =     3.19 MiB ->     2.46 MiB
[  12/ 311] blk.0.ffn_gate.weight                - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[  13/ 311] blk.0.ffn_norm.weight                - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[  14/ 311] blk.0.ffn_up.weight                  - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[  15/ 311] blk.1.attn_k.weight                  - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[  16/ 311] blk.1.attn_k_norm.weight             - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[  17/ 311] blk.1.attn_norm.weight               - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[  18/ 311] blk.1.attn_output.weight             - [  2048,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[  19/ 311] blk.1.attn_q.weight                  - [  1024,   2048,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[  20/ 311] blk.1.attn_q_norm.weight             - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[  21/ 311] blk.1.attn_v.weight                  - [  1024,   1024,      1,      1], type =   q8_0, converting to q6_K .. size =     1.06 MiB ->     0.82 MiB
[  22/ 311] blk.1.ffn_down.weight                - [  3072,   1024,      1,      1], type =   q8_0, converting to q6_K .. size =     3.19 MiB ->     2.46 MiB
[  23/ 311] blk.1.ffn_gate.weight                - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[  24/ 311] blk.1.ffn_norm.weight                - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[  25/ 311] blk.1.ffn_up.weight                  - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[  26/ 311] blk.2.attn_k.weight                  - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[  27/ 311] blk.2.attn_k_norm.weight             - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[  28/ 311] blk.2.attn_norm.weight               - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[  29/ 311] blk.2.attn_output.weight             - [  2048,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[  30/ 311] blk.2.attn_q.weight                  - [  1024,   2048,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[  31/ 311] blk.2.attn_q_norm.weight             - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[  32/ 311] blk.2.attn_v.weight                  - [  1024,   1024,      1,      1], type =   q8_0, converting to q6_K .. size =     1.06 MiB ->     0.82 MiB
[  33/ 311] blk.2.ffn_down.weight                - [  3072,   1024,      1,      1], type =   q8_0, converting to q6_K .. size =     3.19 MiB ->     2.46 MiB
[  34/ 311] blk.2.ffn_gate.weight                - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[  35/ 311] blk.2.ffn_norm.weight                - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[  36/ 311] blk.2.ffn_up.weight                  - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[  37/ 311] blk.3.attn_k.weight                  - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[  38/ 311] blk.3.attn_k_norm.weight             - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[  39/ 311] blk.3.attn_norm.weight               - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[  40/ 311] blk.3.attn_output.weight             - [  2048,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[  41/ 311] blk.3.attn_q.weight                  - [  1024,   2048,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[  42/ 311] blk.3.attn_q_norm.weight             - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[  43/ 311] blk.3.attn_v.weight                  - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[  44/ 311] blk.3.ffn_down.weight                - [  3072,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[  45/ 311] blk.3.ffn_gate.weight                - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[  46/ 311] blk.3.ffn_norm.weight                - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[  47/ 311] blk.3.ffn_up.weight                  - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[  48/ 311] blk.4.attn_k.weight                  - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[  49/ 311] blk.4.attn_k_norm.weight             - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[  50/ 311] blk.4.attn_norm.weight               - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[  51/ 311] blk.4.attn_output.weight             - [  2048,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[  52/ 311] blk.4.attn_q.weight                  - [  1024,   2048,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[  53/ 311] blk.4.attn_q_norm.weight             - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[  54/ 311] blk.4.attn_v.weight                  - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[  55/ 311] blk.4.ffn_down.weight                - [  3072,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[  56/ 311] blk.4.ffn_gate.weight                - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[  57/ 311] blk.4.ffn_norm.weight                - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[  58/ 311] blk.4.ffn_up.weight                  - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[  59/ 311] blk.5.attn_k.weight                  - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[  60/ 311] blk.5.attn_k_norm.weight             - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[  61/ 311] blk.5.attn_norm.weight               - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[  62/ 311] blk.5.attn_output.weight             - [  2048,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[  63/ 311] blk.5.attn_q.weight                  - [  1024,   2048,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[  64/ 311] blk.5.attn_q_norm.weight             - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[  65/ 311] blk.5.attn_v.weight                  - [  1024,   1024,      1,      1], type =   q8_0, converting to q6_K .. size =     1.06 MiB ->     0.82 MiB
[  66/ 311] blk.5.ffn_down.weight                - [  3072,   1024,      1,      1], type =   q8_0, converting to q6_K .. size =     3.19 MiB ->     2.46 MiB
[  67/ 311] blk.5.ffn_gate.weight                - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[  68/ 311] blk.5.ffn_norm.weight                - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[  69/ 311] blk.5.ffn_up.weight                  - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[  70/ 311] blk.6.attn_k.weight                  - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[  71/ 311] blk.6.attn_k_norm.weight             - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[  72/ 311] blk.6.attn_norm.weight               - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[  73/ 311] blk.6.attn_output.weight             - [  2048,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[  74/ 311] blk.6.attn_q.weight                  - [  1024,   2048,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[  75/ 311] blk.6.attn_q_norm.weight             - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[  76/ 311] blk.6.attn_v.weight                  - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[  77/ 311] blk.6.ffn_down.weight                - [  3072,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[  78/ 311] blk.6.ffn_gate.weight                - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[  79/ 311] blk.6.ffn_norm.weight                - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[  80/ 311] blk.6.ffn_up.weight                  - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[  81/ 311] blk.7.attn_k.weight                  - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[  82/ 311] blk.7.attn_k_norm.weight             - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[  83/ 311] blk.7.attn_norm.weight               - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[  84/ 311] blk.7.attn_output.weight             - [  2048,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[  85/ 311] blk.7.attn_q.weight                  - [  1024,   2048,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[  86/ 311] blk.7.attn_q_norm.weight             - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[  87/ 311] blk.7.attn_v.weight                  - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[  88/ 311] blk.7.ffn_down.weight                - [  3072,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[  89/ 311] blk.7.ffn_gate.weight                - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[  90/ 311] blk.7.ffn_norm.weight                - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[  91/ 311] blk.7.ffn_up.weight                  - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[  92/ 311] blk.8.attn_k.weight                  - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[  93/ 311] blk.8.attn_k_norm.weight             - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[  94/ 311] blk.8.attn_norm.weight               - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[  95/ 311] blk.8.attn_output.weight             - [  2048,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[  96/ 311] blk.8.attn_q.weight                  - [  1024,   2048,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[  97/ 311] blk.8.attn_q_norm.weight             - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[  98/ 311] blk.8.attn_v.weight                  - [  1024,   1024,      1,      1], type =   q8_0, converting to q6_K .. size =     1.06 MiB ->     0.82 MiB
[  99/ 311] blk.8.ffn_down.weight                - [  3072,   1024,      1,      1], type =   q8_0, converting to q6_K .. size =     3.19 MiB ->     2.46 MiB
[ 100/ 311] blk.8.ffn_gate.weight                - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 101/ 311] blk.8.ffn_norm.weight                - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 102/ 311] blk.8.ffn_up.weight                  - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 103/ 311] blk.9.attn_k.weight                  - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[ 104/ 311] blk.9.attn_k_norm.weight             - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 105/ 311] blk.9.attn_norm.weight               - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 106/ 311] blk.9.attn_output.weight             - [  2048,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 107/ 311] blk.9.attn_q.weight                  - [  1024,   2048,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 108/ 311] blk.9.attn_q_norm.weight             - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 109/ 311] blk.9.attn_v.weight                  - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[ 110/ 311] blk.9.ffn_down.weight                - [  3072,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 111/ 311] blk.9.ffn_gate.weight                - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 112/ 311] blk.9.ffn_norm.weight                - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 113/ 311] blk.9.ffn_up.weight                  - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 114/ 311] blk.10.attn_k.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[ 115/ 311] blk.10.attn_k_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 116/ 311] blk.10.attn_norm.weight              - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 117/ 311] blk.10.attn_output.weight            - [  2048,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 118/ 311] blk.10.attn_q.weight                 - [  1024,   2048,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 119/ 311] blk.10.attn_q_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 120/ 311] blk.10.attn_v.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[ 121/ 311] blk.10.ffn_down.weight               - [  3072,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 122/ 311] blk.10.ffn_gate.weight               - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 123/ 311] blk.10.ffn_norm.weight               - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 124/ 311] blk.10.ffn_up.weight                 - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 125/ 311] blk.11.attn_k.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[ 126/ 311] blk.11.attn_k_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 127/ 311] blk.11.attn_norm.weight              - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 128/ 311] blk.11.attn_output.weight            - [  2048,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 129/ 311] blk.11.attn_q.weight                 - [  1024,   2048,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 130/ 311] blk.11.attn_q_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 131/ 311] blk.11.attn_v.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q6_K .. size =     1.06 MiB ->     0.82 MiB
[ 132/ 311] blk.11.ffn_down.weight               - [  3072,   1024,      1,      1], type =   q8_0, converting to q6_K .. size =     3.19 MiB ->     2.46 MiB
[ 133/ 311] blk.11.ffn_gate.weight               - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 134/ 311] blk.11.ffn_norm.weight               - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 135/ 311] blk.11.ffn_up.weight                 - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 136/ 311] blk.12.attn_k.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[ 137/ 311] blk.12.attn_k_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 138/ 311] blk.12.attn_norm.weight              - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 139/ 311] blk.12.attn_output.weight            - [  2048,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 140/ 311] blk.12.attn_q.weight                 - [  1024,   2048,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 141/ 311] blk.12.attn_q_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 142/ 311] blk.12.attn_v.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[ 143/ 311] blk.12.ffn_down.weight               - [  3072,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 144/ 311] blk.12.ffn_gate.weight               - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 145/ 311] blk.12.ffn_norm.weight               - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 146/ 311] blk.12.ffn_up.weight                 - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 147/ 311] blk.13.attn_k.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[ 148/ 311] blk.13.attn_k_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 149/ 311] blk.13.attn_norm.weight              - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 150/ 311] blk.13.attn_output.weight            - [  2048,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 151/ 311] blk.13.attn_q.weight                 - [  1024,   2048,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 152/ 311] blk.13.attn_q_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 153/ 311] blk.13.attn_v.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[ 154/ 311] blk.13.ffn_down.weight               - [  3072,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 155/ 311] blk.13.ffn_gate.weight               - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 156/ 311] blk.13.ffn_norm.weight               - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 157/ 311] blk.13.ffn_up.weight                 - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 158/ 311] blk.14.attn_k.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[ 159/ 311] blk.14.attn_k_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 160/ 311] blk.14.attn_norm.weight              - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 161/ 311] blk.14.attn_output.weight            - [  2048,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 162/ 311] blk.14.attn_q.weight                 - [  1024,   2048,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 163/ 311] blk.14.attn_q_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 164/ 311] blk.14.attn_v.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q6_K .. size =     1.06 MiB ->     0.82 MiB
[ 165/ 311] blk.14.ffn_down.weight               - [  3072,   1024,      1,      1], type =   q8_0, converting to q6_K .. size =     3.19 MiB ->     2.46 MiB
[ 166/ 311] blk.14.ffn_gate.weight               - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 167/ 311] blk.14.ffn_norm.weight               - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 168/ 311] blk.14.ffn_up.weight                 - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 169/ 311] blk.15.attn_k.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[ 170/ 311] blk.15.attn_k_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 171/ 311] blk.15.attn_norm.weight              - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 172/ 311] blk.15.attn_output.weight            - [  2048,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 173/ 311] blk.15.attn_q.weight                 - [  1024,   2048,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 174/ 311] blk.15.attn_q_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 175/ 311] blk.15.attn_v.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[ 176/ 311] blk.15.ffn_down.weight               - [  3072,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 177/ 311] blk.15.ffn_gate.weight               - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 178/ 311] blk.15.ffn_norm.weight               - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 179/ 311] blk.15.ffn_up.weight                 - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 180/ 311] blk.16.attn_k.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[ 181/ 311] blk.16.attn_k_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 182/ 311] blk.16.attn_norm.weight              - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 183/ 311] blk.16.attn_output.weight            - [  2048,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 184/ 311] blk.16.attn_q.weight                 - [  1024,   2048,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 185/ 311] blk.16.attn_q_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 186/ 311] blk.16.attn_v.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[ 187/ 311] blk.16.ffn_down.weight               - [  3072,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 188/ 311] blk.16.ffn_gate.weight               - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 189/ 311] blk.16.ffn_norm.weight               - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 190/ 311] blk.16.ffn_up.weight                 - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 191/ 311] blk.17.attn_k.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[ 192/ 311] blk.17.attn_k_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 193/ 311] blk.17.attn_norm.weight              - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 194/ 311] blk.17.attn_output.weight            - [  2048,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 195/ 311] blk.17.attn_q.weight                 - [  1024,   2048,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 196/ 311] blk.17.attn_q_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 197/ 311] blk.17.attn_v.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q6_K .. size =     1.06 MiB ->     0.82 MiB
[ 198/ 311] blk.17.ffn_down.weight               - [  3072,   1024,      1,      1], type =   q8_0, converting to q6_K .. size =     3.19 MiB ->     2.46 MiB
[ 199/ 311] blk.17.ffn_gate.weight               - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 200/ 311] blk.17.ffn_norm.weight               - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 201/ 311] blk.17.ffn_up.weight                 - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 202/ 311] blk.18.attn_k.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[ 203/ 311] blk.18.attn_k_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 204/ 311] blk.18.attn_norm.weight              - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 205/ 311] blk.18.attn_output.weight            - [  2048,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 206/ 311] blk.18.attn_q.weight                 - [  1024,   2048,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 207/ 311] blk.18.attn_q_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 208/ 311] blk.18.attn_v.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[ 209/ 311] blk.18.ffn_down.weight               - [  3072,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 210/ 311] blk.18.ffn_gate.weight               - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 211/ 311] blk.18.ffn_norm.weight               - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 212/ 311] blk.18.ffn_up.weight                 - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 213/ 311] blk.19.attn_k.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[ 214/ 311] blk.19.attn_k_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 215/ 311] blk.19.attn_norm.weight              - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 216/ 311] blk.19.attn_output.weight            - [  2048,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 217/ 311] blk.19.attn_q.weight                 - [  1024,   2048,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 218/ 311] blk.19.attn_q_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 219/ 311] blk.19.attn_v.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[ 220/ 311] blk.19.ffn_down.weight               - [  3072,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 221/ 311] blk.19.ffn_gate.weight               - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 222/ 311] blk.19.ffn_norm.weight               - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 223/ 311] blk.19.ffn_up.weight                 - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 224/ 311] blk.20.attn_k.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[ 225/ 311] blk.20.attn_k_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 226/ 311] blk.20.attn_norm.weight              - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 227/ 311] blk.20.attn_output.weight            - [  2048,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 228/ 311] blk.20.attn_q.weight                 - [  1024,   2048,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 229/ 311] blk.20.attn_q_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 230/ 311] blk.20.attn_v.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q6_K .. size =     1.06 MiB ->     0.82 MiB
[ 231/ 311] blk.20.ffn_down.weight               - [  3072,   1024,      1,      1], type =   q8_0, converting to q6_K .. size =     3.19 MiB ->     2.46 MiB
[ 232/ 311] blk.20.ffn_gate.weight               - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 233/ 311] blk.20.ffn_norm.weight               - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 234/ 311] blk.20.ffn_up.weight                 - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 235/ 311] blk.21.attn_k.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[ 236/ 311] blk.21.attn_k_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 237/ 311] blk.21.attn_norm.weight              - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 238/ 311] blk.21.attn_output.weight            - [  2048,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 239/ 311] blk.21.attn_q.weight                 - [  1024,   2048,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 240/ 311] blk.21.attn_q_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 241/ 311] blk.21.attn_v.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[ 242/ 311] blk.21.ffn_down.weight               - [  3072,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 243/ 311] blk.21.ffn_gate.weight               - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 244/ 311] blk.21.ffn_norm.weight               - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 245/ 311] blk.21.ffn_up.weight                 - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 246/ 311] blk.22.attn_k.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[ 247/ 311] blk.22.attn_k_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 248/ 311] blk.22.attn_norm.weight              - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 249/ 311] blk.22.attn_output.weight            - [  2048,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 250/ 311] blk.22.attn_q.weight                 - [  1024,   2048,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 251/ 311] blk.22.attn_q_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 252/ 311] blk.22.attn_v.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[ 253/ 311] blk.22.ffn_down.weight               - [  3072,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 254/ 311] blk.22.ffn_gate.weight               - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 255/ 311] blk.22.ffn_norm.weight               - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 256/ 311] blk.22.ffn_up.weight                 - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 257/ 311] blk.23.attn_k.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[ 258/ 311] blk.23.attn_k_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 259/ 311] blk.23.attn_norm.weight              - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 260/ 311] blk.23.attn_output.weight            - [  2048,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 261/ 311] blk.23.attn_q.weight                 - [  1024,   2048,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 262/ 311] blk.23.attn_q_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 263/ 311] blk.23.attn_v.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q6_K .. size =     1.06 MiB ->     0.82 MiB
[ 264/ 311] blk.23.ffn_down.weight               - [  3072,   1024,      1,      1], type =   q8_0, converting to q6_K .. size =     3.19 MiB ->     2.46 MiB
[ 265/ 311] blk.23.ffn_gate.weight               - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 266/ 311] blk.23.ffn_norm.weight               - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 267/ 311] blk.23.ffn_up.weight                 - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 268/ 311] blk.24.attn_k.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[ 269/ 311] blk.24.attn_k_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 270/ 311] blk.24.attn_norm.weight              - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 271/ 311] blk.24.attn_output.weight            - [  2048,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 272/ 311] blk.24.attn_q.weight                 - [  1024,   2048,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 273/ 311] blk.24.attn_q_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 274/ 311] blk.24.attn_v.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q6_K .. size =     1.06 MiB ->     0.82 MiB
[ 275/ 311] blk.24.ffn_down.weight               - [  3072,   1024,      1,      1], type =   q8_0, converting to q6_K .. size =     3.19 MiB ->     2.46 MiB
[ 276/ 311] blk.24.ffn_gate.weight               - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 277/ 311] blk.24.ffn_norm.weight               - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 278/ 311] blk.24.ffn_up.weight                 - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 279/ 311] blk.25.attn_k.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[ 280/ 311] blk.25.attn_k_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 281/ 311] blk.25.attn_norm.weight              - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 282/ 311] blk.25.attn_output.weight            - [  2048,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 283/ 311] blk.25.attn_q.weight                 - [  1024,   2048,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 284/ 311] blk.25.attn_q_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 285/ 311] blk.25.attn_v.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q6_K .. size =     1.06 MiB ->     0.82 MiB
[ 286/ 311] blk.25.ffn_down.weight               - [  3072,   1024,      1,      1], type =   q8_0, converting to q6_K .. size =     3.19 MiB ->     2.46 MiB
[ 287/ 311] blk.25.ffn_gate.weight               - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 288/ 311] blk.25.ffn_norm.weight               - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 289/ 311] blk.25.ffn_up.weight                 - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 290/ 311] blk.26.attn_k.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[ 291/ 311] blk.26.attn_k_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 292/ 311] blk.26.attn_norm.weight              - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 293/ 311] blk.26.attn_output.weight            - [  2048,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 294/ 311] blk.26.attn_q.weight                 - [  1024,   2048,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 295/ 311] blk.26.attn_q_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 296/ 311] blk.26.attn_v.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q6_K .. size =     1.06 MiB ->     0.82 MiB
[ 297/ 311] blk.26.ffn_down.weight               - [  3072,   1024,      1,      1], type =   q8_0, converting to q6_K .. size =     3.19 MiB ->     2.46 MiB
[ 298/ 311] blk.26.ffn_gate.weight               - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 299/ 311] blk.26.ffn_norm.weight               - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 300/ 311] blk.26.ffn_up.weight                 - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 301/ 311] blk.27.attn_k.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[ 302/ 311] blk.27.attn_k_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 303/ 311] blk.27.attn_norm.weight              - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 304/ 311] blk.27.attn_output.weight            - [  2048,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 305/ 311] blk.27.attn_q.weight                 - [  1024,   2048,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 306/ 311] blk.27.attn_q_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 307/ 311] blk.27.attn_v.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q6_K .. size =     1.06 MiB ->     0.82 MiB
[ 308/ 311] blk.27.ffn_down.weight               - [  3072,   1024,      1,      1], type =   q8_0, converting to q6_K .. size =     3.19 MiB ->     2.46 MiB
[ 309/ 311] blk.27.ffn_gate.weight               - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 310/ 311] blk.27.ffn_norm.weight               - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 311/ 311] blk.27.ffn_up.weight                 - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
llama_model_quantize_impl: model size  =   761.80 MiB (8.50 BPW)
llama_model_quantize_impl: quant size  =   456.11 MiB (5.09 BPW)

llama_quantize: quantize time =  5211.17 ms
llama_quantize:    total time =  5211.17 ms
+ echo PASS
PASS
+ echo

+ /root/llama.me/build-ci-release/bin/llama-completion -no-cnv --model /root/llama.me/tmp/mnt/models/quantize/ggml-model-requant-00001-of-00012.gguf -p 'I believe the meaning of life is' --n-predict 32
0.00.001.812 I llama_completion: llama backend init
0.00.001.818 I llama_completion: load the model and apply lora adapter, if any
0.00.001.826 I common_init_result: fitting params to device memory ...
0.00.001.826 I common_init_result: (for bugs during this step try to reproduce them with -fit off, or provide --verbose logs if the bug only occurs with -fit on)
0.00.185.525 I common_params_fit_impl: projected to use 5032 MiB of host memory vs. 77898 MiB of total host memory
0.00.267.524 W load: control-looking token: 128247 '</s>' was not control-type; this is probably a bug in the model. its type will be overridden
0.02.092.503 I common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
0.02.164.145 I llama_completion: llama threadpool init, n_threads = 8
0.02.164.163 I 
0.02.164.220 I system_info: n_threads = 8 (n_threads_batch = 8) / 16 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 | 
0.02.164.222 I 
0.02.164.294 I sampler seed: 615738873
0.02.164.301 I sampler params: 
	repeat_last_n = 64, repeat_penalty = 1,000, frequency_penalty = 0,000, presence_penalty = 0,000
	dry_multiplier = 0,000, dry_base = 1,750, dry_allowed_length = 2, dry_penalty_last_n = -1
	top_k = 40, top_p = 0,950, min_p = 0,050, xtc_probability = 0,000, xtc_threshold = 0,100, typical_p = 1,000, top_n_sigma = -1,000, temp = 0,800
	mirostat = 0, mirostat_lr = 0,100, mirostat_ent = 5,000, adaptive_target = -1,000, adaptive_decay = 0,900
0.02.164.306 I sampler chain: logits -> ?penalties -> ?dry -> ?top-n-sigma -> top-k -> ?typical -> top-p -> min-p -> ?xtc -> temp-ext -> dist 
0.02.164.306 I generate: n_ctx = 40960, n_batch = 2048, n_predict = 32, n_keep = 0
0.02.164.306 I 
I believe the meaning of life is to live in the present and to accept the present moment. If I had to summarize this belief, I would say it is the result of the interplay between

0.02.432.326 I common_perf_print:    sampling time =       4,71 ms
0.02.432.327 I common_perf_print:    samplers time =       1,64 ms /    39 tokens
0.02.432.331 I common_perf_print:        load time =    1939,94 ms
0.02.432.332 I common_perf_print: prompt eval time =      14,94 ms /     7 tokens (    2,13 ms per token,   468,60 tokens per second)
0.02.432.346 I common_perf_print:        eval time =     247,69 ms /    31 runs   (    7,99 ms per token,   125,16 tokens per second)
0.02.432.347 I common_perf_print:       total time =     268,19 ms /    38 tokens
0.02.432.348 I common_perf_print: unaccounted time =       0,85 ms /   0,3 %      (total - sampling - prompt eval - eval) / (total)
0.02.432.348 I common_perf_print:    graphs reused =         30
+ echo PASS
PASS
+ echo

+ /root/llama.me/build-ci-release/bin/llama-quantize --allow-requantize /root/llama.me/tmp/mnt/models/quantize/ggml-model-split-00001-of-00012.gguf /root/llama.me/tmp/mnt/models/quantize/ggml-model-requant-merge.gguf Q4_K
llama_print_build_info: build = 9440 (eb91ed029)
llama_print_build_info: built with GNU 13.3.0 for Linux x86_64
llama_quantize: quantizing '/root/llama.me/tmp/mnt/models/quantize/ggml-model-split-00001-of-00012.gguf' to '/root/llama.me/tmp/mnt/models/quantize/ggml-model-requant-merge.gguf' as Q4_K
llama_model_loader: additional 11 GGUFs metadata loaded.
llama_model_loader: loaded meta data with 37 key-value pairs and 311 tensors from /root/llama.me/tmp/mnt/models/quantize/ggml-model-split-00001-of-00012.gguf (version GGUF V3 (latest))
llama_model_loader: Dumping metadata keys/values. Note: KV overrides do not apply in this output.
llama_model_loader: - kv   0:                       general.architecture str              = qwen3
llama_model_loader: - kv   1:                               general.type str              = model
llama_model_loader: - kv   2:                               general.name str              = Qwen3 0.6B
llama_model_loader: - kv   3:                           general.basename str              = Qwen3
llama_model_loader: - kv   4:                         general.size_label str              = 0.6B
llama_model_loader: - kv   5:                            general.license str              = apache-2.0
llama_model_loader: - kv   6:                       general.license.link str              = https://huggingface.co/Qwen/Qwen3-0.6...
llama_model_loader: - kv   7:                   general.base_model.count u32              = 1
llama_model_loader: - kv   8:                  general.base_model.0.name str              = Qwen3 0.6B Base
llama_model_loader: - kv   9:          general.base_model.0.organization str              = Qwen
llama_model_loader: - kv  10:              general.base_model.0.repo_url str              = https://huggingface.co/Qwen/Qwen3-0.6...
llama_model_loader: - kv  11:                               general.tags arr[str,1]       = ["text-generation"]
llama_model_loader: - kv  12:                          qwen3.block_count u32              = 28
llama_model_loader: - kv  13:                       qwen3.context_length u32              = 40960
llama_model_loader: - kv  14:                     qwen3.embedding_length u32              = 1024
llama_model_loader: - kv  15:                  qwen3.feed_forward_length u32              = 3072
llama_model_loader: - kv  16:                 qwen3.attention.head_count u32              = 16
llama_model_loader: - kv  17:              qwen3.attention.head_count_kv u32              = 8
llama_model_loader: - kv  18:                       qwen3.rope.freq_base f32              = 1000000.000000
llama_model_loader: - kv  19:     qwen3.attention.layer_norm_rms_epsilon f32              = 0.000001
llama_model_loader: - kv  20:                 qwen3.attention.key_length u32              = 128
llama_model_loader: - kv  21:               qwen3.attention.value_length u32              = 128
llama_model_loader: - kv  22:                       tokenizer.ggml.model str              = gpt2
llama_model_loader: - kv  23:                         tokenizer.ggml.pre str              = qwen2
llama_model_loader: - kv  24:                      tokenizer.ggml.tokens arr[str,151936]  = ["!", "\"", "#", "$", "%", "&", "'", ...
llama_model_loader: - kv  25:                  tokenizer.ggml.token_type arr[i32,151936]  = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, ...
llama_model_loader: - kv  26:                      tokenizer.ggml.merges arr[str,151387]  = ["Ġ Ġ", "ĠĠ ĠĠ", "i n", "Ġ t",...
llama_model_loader: - kv  27:                tokenizer.ggml.eos_token_id u32              = 151645
llama_model_loader: - kv  28:            tokenizer.ggml.padding_token_id u32              = 151643
llama_model_loader: - kv  29:                tokenizer.ggml.bos_token_id u32              = 151643
llama_model_loader: - kv  30:               tokenizer.ggml.add_bos_token bool             = false
llama_model_loader: - kv  31:                    tokenizer.chat_template str              = {%- if tools %}\n    {{- '<|im_start|>...
llama_model_loader: - kv  32:               general.quantization_version u32              = 2
llama_model_loader: - kv  33:                          general.file_type u32              = 7
llama_model_loader: - kv  34:                                   split.no u16              = 0
llama_model_loader: - kv  35:                        split.tensors.count i32              = 311
llama_model_loader: - kv  36:                                split.count u16              = 12
llama_model_loader: - type  f32:  113 tensors
llama_model_loader: - type q8_0:  198 tensors
[   1/ 311] output.weight                        - [  1024, 151936,      1,      1], type =   q8_0, converting to q6_K .. size =   157.65 MiB ->   121.71 MiB
[   2/ 311] output_norm.weight                   - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[   3/ 311] token_embd.weight                    - [  1024, 151936,      1,      1], type =   q8_0, converting to q4_K .. size =   157.65 MiB ->    83.46 MiB
[   4/ 311] blk.0.attn_k.weight                  - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[   5/ 311] blk.0.attn_k_norm.weight             - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[   6/ 311] blk.0.attn_norm.weight               - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[   7/ 311] blk.0.attn_output.weight             - [  2048,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[   8/ 311] blk.0.attn_q.weight                  - [  1024,   2048,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[   9/ 311] blk.0.attn_q_norm.weight             - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[  10/ 311] blk.0.attn_v.weight                  - [  1024,   1024,      1,      1], type =   q8_0, converting to q6_K .. size =     1.06 MiB ->     0.82 MiB
[  11/ 311] blk.0.ffn_down.weight                - [  3072,   1024,      1,      1], type =   q8_0, converting to q6_K .. size =     3.19 MiB ->     2.46 MiB
[  12/ 311] blk.0.ffn_gate.weight                - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[  13/ 311] blk.0.ffn_norm.weight                - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[  14/ 311] blk.0.ffn_up.weight                  - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[  15/ 311] blk.1.attn_k.weight                  - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[  16/ 311] blk.1.attn_k_norm.weight             - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[  17/ 311] blk.1.attn_norm.weight               - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[  18/ 311] blk.1.attn_output.weight             - [  2048,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[  19/ 311] blk.1.attn_q.weight                  - [  1024,   2048,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[  20/ 311] blk.1.attn_q_norm.weight             - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[  21/ 311] blk.1.attn_v.weight                  - [  1024,   1024,      1,      1], type =   q8_0, converting to q6_K .. size =     1.06 MiB ->     0.82 MiB
[  22/ 311] blk.1.ffn_down.weight                - [  3072,   1024,      1,      1], type =   q8_0, converting to q6_K .. size =     3.19 MiB ->     2.46 MiB
[  23/ 311] blk.1.ffn_gate.weight                - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[  24/ 311] blk.1.ffn_norm.weight                - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[  25/ 311] blk.1.ffn_up.weight                  - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[  26/ 311] blk.2.attn_k.weight                  - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[  27/ 311] blk.2.attn_k_norm.weight             - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[  28/ 311] blk.2.attn_norm.weight               - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[  29/ 311] blk.2.attn_output.weight             - [  2048,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[  30/ 311] blk.2.attn_q.weight                  - [  1024,   2048,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[  31/ 311] blk.2.attn_q_norm.weight             - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[  32/ 311] blk.2.attn_v.weight                  - [  1024,   1024,      1,      1], type =   q8_0, converting to q6_K .. size =     1.06 MiB ->     0.82 MiB
[  33/ 311] blk.2.ffn_down.weight                - [  3072,   1024,      1,      1], type =   q8_0, converting to q6_K .. size =     3.19 MiB ->     2.46 MiB
[  34/ 311] blk.2.ffn_gate.weight                - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[  35/ 311] blk.2.ffn_norm.weight                - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[  36/ 311] blk.2.ffn_up.weight                  - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[  37/ 311] blk.3.attn_k.weight                  - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[  38/ 311] blk.3.attn_k_norm.weight             - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[  39/ 311] blk.3.attn_norm.weight               - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[  40/ 311] blk.3.attn_output.weight             - [  2048,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[  41/ 311] blk.3.attn_q.weight                  - [  1024,   2048,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[  42/ 311] blk.3.attn_q_norm.weight             - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[  43/ 311] blk.3.attn_v.weight                  - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[  44/ 311] blk.3.ffn_down.weight                - [  3072,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[  45/ 311] blk.3.ffn_gate.weight                - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[  46/ 311] blk.3.ffn_norm.weight                - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[  47/ 311] blk.3.ffn_up.weight                  - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[  48/ 311] blk.4.attn_k.weight                  - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[  49/ 311] blk.4.attn_k_norm.weight             - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[  50/ 311] blk.4.attn_norm.weight               - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[  51/ 311] blk.4.attn_output.weight             - [  2048,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[  52/ 311] blk.4.attn_q.weight                  - [  1024,   2048,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[  53/ 311] blk.4.attn_q_norm.weight             - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[  54/ 311] blk.4.attn_v.weight                  - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[  55/ 311] blk.4.ffn_down.weight                - [  3072,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[  56/ 311] blk.4.ffn_gate.weight                - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[  57/ 311] blk.4.ffn_norm.weight                - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[  58/ 311] blk.4.ffn_up.weight                  - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[  59/ 311] blk.5.attn_k.weight                  - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[  60/ 311] blk.5.attn_k_norm.weight             - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[  61/ 311] blk.5.attn_norm.weight               - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[  62/ 311] blk.5.attn_output.weight             - [  2048,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[  63/ 311] blk.5.attn_q.weight                  - [  1024,   2048,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[  64/ 311] blk.5.attn_q_norm.weight             - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[  65/ 311] blk.5.attn_v.weight                  - [  1024,   1024,      1,      1], type =   q8_0, converting to q6_K .. size =     1.06 MiB ->     0.82 MiB
[  66/ 311] blk.5.ffn_down.weight                - [  3072,   1024,      1,      1], type =   q8_0, converting to q6_K .. size =     3.19 MiB ->     2.46 MiB
[  67/ 311] blk.5.ffn_gate.weight                - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[  68/ 311] blk.5.ffn_norm.weight                - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[  69/ 311] blk.5.ffn_up.weight                  - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[  70/ 311] blk.6.attn_k.weight                  - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[  71/ 311] blk.6.attn_k_norm.weight             - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[  72/ 311] blk.6.attn_norm.weight               - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[  73/ 311] blk.6.attn_output.weight             - [  2048,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[  74/ 311] blk.6.attn_q.weight                  - [  1024,   2048,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[  75/ 311] blk.6.attn_q_norm.weight             - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[  76/ 311] blk.6.attn_v.weight                  - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[  77/ 311] blk.6.ffn_down.weight                - [  3072,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[  78/ 311] blk.6.ffn_gate.weight                - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[  79/ 311] blk.6.ffn_norm.weight                - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[  80/ 311] blk.6.ffn_up.weight                  - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[  81/ 311] blk.7.attn_k.weight                  - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[  82/ 311] blk.7.attn_k_norm.weight             - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[  83/ 311] blk.7.attn_norm.weight               - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[  84/ 311] blk.7.attn_output.weight             - [  2048,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[  85/ 311] blk.7.attn_q.weight                  - [  1024,   2048,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[  86/ 311] blk.7.attn_q_norm.weight             - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[  87/ 311] blk.7.attn_v.weight                  - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[  88/ 311] blk.7.ffn_down.weight                - [  3072,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[  89/ 311] blk.7.ffn_gate.weight                - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[  90/ 311] blk.7.ffn_norm.weight                - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[  91/ 311] blk.7.ffn_up.weight                  - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[  92/ 311] blk.8.attn_k.weight                  - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[  93/ 311] blk.8.attn_k_norm.weight             - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[  94/ 311] blk.8.attn_norm.weight               - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[  95/ 311] blk.8.attn_output.weight             - [  2048,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[  96/ 311] blk.8.attn_q.weight                  - [  1024,   2048,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[  97/ 311] blk.8.attn_q_norm.weight             - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[  98/ 311] blk.8.attn_v.weight                  - [  1024,   1024,      1,      1], type =   q8_0, converting to q6_K .. size =     1.06 MiB ->     0.82 MiB
[  99/ 311] blk.8.ffn_down.weight                - [  3072,   1024,      1,      1], type =   q8_0, converting to q6_K .. size =     3.19 MiB ->     2.46 MiB
[ 100/ 311] blk.8.ffn_gate.weight                - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 101/ 311] blk.8.ffn_norm.weight                - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 102/ 311] blk.8.ffn_up.weight                  - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 103/ 311] blk.9.attn_k.weight                  - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[ 104/ 311] blk.9.attn_k_norm.weight             - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 105/ 311] blk.9.attn_norm.weight               - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 106/ 311] blk.9.attn_output.weight             - [  2048,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 107/ 311] blk.9.attn_q.weight                  - [  1024,   2048,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 108/ 311] blk.9.attn_q_norm.weight             - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 109/ 311] blk.9.attn_v.weight                  - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[ 110/ 311] blk.9.ffn_down.weight                - [  3072,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 111/ 311] blk.9.ffn_gate.weight                - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 112/ 311] blk.9.ffn_norm.weight                - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 113/ 311] blk.9.ffn_up.weight                  - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 114/ 311] blk.10.attn_k.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[ 115/ 311] blk.10.attn_k_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 116/ 311] blk.10.attn_norm.weight              - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 117/ 311] blk.10.attn_output.weight            - [  2048,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 118/ 311] blk.10.attn_q.weight                 - [  1024,   2048,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 119/ 311] blk.10.attn_q_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 120/ 311] blk.10.attn_v.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[ 121/ 311] blk.10.ffn_down.weight               - [  3072,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 122/ 311] blk.10.ffn_gate.weight               - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 123/ 311] blk.10.ffn_norm.weight               - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 124/ 311] blk.10.ffn_up.weight                 - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 125/ 311] blk.11.attn_k.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[ 126/ 311] blk.11.attn_k_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 127/ 311] blk.11.attn_norm.weight              - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 128/ 311] blk.11.attn_output.weight            - [  2048,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 129/ 311] blk.11.attn_q.weight                 - [  1024,   2048,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 130/ 311] blk.11.attn_q_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 131/ 311] blk.11.attn_v.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q6_K .. size =     1.06 MiB ->     0.82 MiB
[ 132/ 311] blk.11.ffn_down.weight               - [  3072,   1024,      1,      1], type =   q8_0, converting to q6_K .. size =     3.19 MiB ->     2.46 MiB
[ 133/ 311] blk.11.ffn_gate.weight               - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 134/ 311] blk.11.ffn_norm.weight               - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 135/ 311] blk.11.ffn_up.weight                 - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 136/ 311] blk.12.attn_k.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[ 137/ 311] blk.12.attn_k_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 138/ 311] blk.12.attn_norm.weight              - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 139/ 311] blk.12.attn_output.weight            - [  2048,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 140/ 311] blk.12.attn_q.weight                 - [  1024,   2048,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 141/ 311] blk.12.attn_q_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 142/ 311] blk.12.attn_v.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[ 143/ 311] blk.12.ffn_down.weight               - [  3072,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 144/ 311] blk.12.ffn_gate.weight               - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 145/ 311] blk.12.ffn_norm.weight               - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 146/ 311] blk.12.ffn_up.weight                 - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 147/ 311] blk.13.attn_k.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[ 148/ 311] blk.13.attn_k_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 149/ 311] blk.13.attn_norm.weight              - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 150/ 311] blk.13.attn_output.weight            - [  2048,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 151/ 311] blk.13.attn_q.weight                 - [  1024,   2048,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 152/ 311] blk.13.attn_q_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 153/ 311] blk.13.attn_v.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[ 154/ 311] blk.13.ffn_down.weight               - [  3072,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 155/ 311] blk.13.ffn_gate.weight               - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 156/ 311] blk.13.ffn_norm.weight               - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 157/ 311] blk.13.ffn_up.weight                 - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 158/ 311] blk.14.attn_k.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[ 159/ 311] blk.14.attn_k_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 160/ 311] blk.14.attn_norm.weight              - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 161/ 311] blk.14.attn_output.weight            - [  2048,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 162/ 311] blk.14.attn_q.weight                 - [  1024,   2048,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 163/ 311] blk.14.attn_q_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 164/ 311] blk.14.attn_v.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q6_K .. size =     1.06 MiB ->     0.82 MiB
[ 165/ 311] blk.14.ffn_down.weight               - [  3072,   1024,      1,      1], type =   q8_0, converting to q6_K .. size =     3.19 MiB ->     2.46 MiB
[ 166/ 311] blk.14.ffn_gate.weight               - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 167/ 311] blk.14.ffn_norm.weight               - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 168/ 311] blk.14.ffn_up.weight                 - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 169/ 311] blk.15.attn_k.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[ 170/ 311] blk.15.attn_k_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 171/ 311] blk.15.attn_norm.weight              - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 172/ 311] blk.15.attn_output.weight            - [  2048,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 173/ 311] blk.15.attn_q.weight                 - [  1024,   2048,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 174/ 311] blk.15.attn_q_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 175/ 311] blk.15.attn_v.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[ 176/ 311] blk.15.ffn_down.weight               - [  3072,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 177/ 311] blk.15.ffn_gate.weight               - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 178/ 311] blk.15.ffn_norm.weight               - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 179/ 311] blk.15.ffn_up.weight                 - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 180/ 311] blk.16.attn_k.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[ 181/ 311] blk.16.attn_k_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 182/ 311] blk.16.attn_norm.weight              - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 183/ 311] blk.16.attn_output.weight            - [  2048,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 184/ 311] blk.16.attn_q.weight                 - [  1024,   2048,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 185/ 311] blk.16.attn_q_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 186/ 311] blk.16.attn_v.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[ 187/ 311] blk.16.ffn_down.weight               - [  3072,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 188/ 311] blk.16.ffn_gate.weight               - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 189/ 311] blk.16.ffn_norm.weight               - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 190/ 311] blk.16.ffn_up.weight                 - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 191/ 311] blk.17.attn_k.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[ 192/ 311] blk.17.attn_k_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 193/ 311] blk.17.attn_norm.weight              - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 194/ 311] blk.17.attn_output.weight            - [  2048,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 195/ 311] blk.17.attn_q.weight                 - [  1024,   2048,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 196/ 311] blk.17.attn_q_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 197/ 311] blk.17.attn_v.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q6_K .. size =     1.06 MiB ->     0.82 MiB
[ 198/ 311] blk.17.ffn_down.weight               - [  3072,   1024,      1,      1], type =   q8_0, converting to q6_K .. size =     3.19 MiB ->     2.46 MiB
[ 199/ 311] blk.17.ffn_gate.weight               - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 200/ 311] blk.17.ffn_norm.weight               - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 201/ 311] blk.17.ffn_up.weight                 - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 202/ 311] blk.18.attn_k.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[ 203/ 311] blk.18.attn_k_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 204/ 311] blk.18.attn_norm.weight              - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 205/ 311] blk.18.attn_output.weight            - [  2048,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 206/ 311] blk.18.attn_q.weight                 - [  1024,   2048,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 207/ 311] blk.18.attn_q_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 208/ 311] blk.18.attn_v.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[ 209/ 311] blk.18.ffn_down.weight               - [  3072,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 210/ 311] blk.18.ffn_gate.weight               - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 211/ 311] blk.18.ffn_norm.weight               - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 212/ 311] blk.18.ffn_up.weight                 - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 213/ 311] blk.19.attn_k.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[ 214/ 311] blk.19.attn_k_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 215/ 311] blk.19.attn_norm.weight              - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 216/ 311] blk.19.attn_output.weight            - [  2048,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 217/ 311] blk.19.attn_q.weight                 - [  1024,   2048,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 218/ 311] blk.19.attn_q_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 219/ 311] blk.19.attn_v.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[ 220/ 311] blk.19.ffn_down.weight               - [  3072,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 221/ 311] blk.19.ffn_gate.weight               - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 222/ 311] blk.19.ffn_norm.weight               - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 223/ 311] blk.19.ffn_up.weight                 - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 224/ 311] blk.20.attn_k.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[ 225/ 311] blk.20.attn_k_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 226/ 311] blk.20.attn_norm.weight              - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 227/ 311] blk.20.attn_output.weight            - [  2048,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 228/ 311] blk.20.attn_q.weight                 - [  1024,   2048,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 229/ 311] blk.20.attn_q_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 230/ 311] blk.20.attn_v.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q6_K .. size =     1.06 MiB ->     0.82 MiB
[ 231/ 311] blk.20.ffn_down.weight               - [  3072,   1024,      1,      1], type =   q8_0, converting to q6_K .. size =     3.19 MiB ->     2.46 MiB
[ 232/ 311] blk.20.ffn_gate.weight               - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 233/ 311] blk.20.ffn_norm.weight               - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 234/ 311] blk.20.ffn_up.weight                 - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 235/ 311] blk.21.attn_k.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[ 236/ 311] blk.21.attn_k_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 237/ 311] blk.21.attn_norm.weight              - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 238/ 311] blk.21.attn_output.weight            - [  2048,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 239/ 311] blk.21.attn_q.weight                 - [  1024,   2048,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 240/ 311] blk.21.attn_q_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 241/ 311] blk.21.attn_v.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[ 242/ 311] blk.21.ffn_down.weight               - [  3072,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 243/ 311] blk.21.ffn_gate.weight               - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 244/ 311] blk.21.ffn_norm.weight               - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 245/ 311] blk.21.ffn_up.weight                 - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 246/ 311] blk.22.attn_k.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[ 247/ 311] blk.22.attn_k_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 248/ 311] blk.22.attn_norm.weight              - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 249/ 311] blk.22.attn_output.weight            - [  2048,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 250/ 311] blk.22.attn_q.weight                 - [  1024,   2048,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 251/ 311] blk.22.attn_q_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 252/ 311] blk.22.attn_v.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[ 253/ 311] blk.22.ffn_down.weight               - [  3072,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 254/ 311] blk.22.ffn_gate.weight               - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 255/ 311] blk.22.ffn_norm.weight               - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 256/ 311] blk.22.ffn_up.weight                 - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 257/ 311] blk.23.attn_k.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[ 258/ 311] blk.23.attn_k_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 259/ 311] blk.23.attn_norm.weight              - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 260/ 311] blk.23.attn_output.weight            - [  2048,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 261/ 311] blk.23.attn_q.weight                 - [  1024,   2048,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 262/ 311] blk.23.attn_q_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 263/ 311] blk.23.attn_v.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q6_K .. size =     1.06 MiB ->     0.82 MiB
[ 264/ 311] blk.23.ffn_down.weight               - [  3072,   1024,      1,      1], type =   q8_0, converting to q6_K .. size =     3.19 MiB ->     2.46 MiB
[ 265/ 311] blk.23.ffn_gate.weight               - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 266/ 311] blk.23.ffn_norm.weight               - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 267/ 311] blk.23.ffn_up.weight                 - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 268/ 311] blk.24.attn_k.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[ 269/ 311] blk.24.attn_k_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 270/ 311] blk.24.attn_norm.weight              - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 271/ 311] blk.24.attn_output.weight            - [  2048,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 272/ 311] blk.24.attn_q.weight                 - [  1024,   2048,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 273/ 311] blk.24.attn_q_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 274/ 311] blk.24.attn_v.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q6_K .. size =     1.06 MiB ->     0.82 MiB
[ 275/ 311] blk.24.ffn_down.weight               - [  3072,   1024,      1,      1], type =   q8_0, converting to q6_K .. size =     3.19 MiB ->     2.46 MiB
[ 276/ 311] blk.24.ffn_gate.weight               - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 277/ 311] blk.24.ffn_norm.weight               - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 278/ 311] blk.24.ffn_up.weight                 - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 279/ 311] blk.25.attn_k.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[ 280/ 311] blk.25.attn_k_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 281/ 311] blk.25.attn_norm.weight              - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 282/ 311] blk.25.attn_output.weight            - [  2048,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 283/ 311] blk.25.attn_q.weight                 - [  1024,   2048,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 284/ 311] blk.25.attn_q_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 285/ 311] blk.25.attn_v.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q6_K .. size =     1.06 MiB ->     0.82 MiB
[ 286/ 311] blk.25.ffn_down.weight               - [  3072,   1024,      1,      1], type =   q8_0, converting to q6_K .. size =     3.19 MiB ->     2.46 MiB
[ 287/ 311] blk.25.ffn_gate.weight               - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 288/ 311] blk.25.ffn_norm.weight               - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 289/ 311] blk.25.ffn_up.weight                 - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 290/ 311] blk.26.attn_k.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[ 291/ 311] blk.26.attn_k_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 292/ 311] blk.26.attn_norm.weight              - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 293/ 311] blk.26.attn_output.weight            - [  2048,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 294/ 311] blk.26.attn_q.weight                 - [  1024,   2048,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 295/ 311] blk.26.attn_q_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 296/ 311] blk.26.attn_v.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q6_K .. size =     1.06 MiB ->     0.82 MiB
[ 297/ 311] blk.26.ffn_down.weight               - [  3072,   1024,      1,      1], type =   q8_0, converting to q6_K .. size =     3.19 MiB ->     2.46 MiB
[ 298/ 311] blk.26.ffn_gate.weight               - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 299/ 311] blk.26.ffn_norm.weight               - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 300/ 311] blk.26.ffn_up.weight                 - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 301/ 311] blk.27.attn_k.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     1.06 MiB ->     0.56 MiB
[ 302/ 311] blk.27.attn_k_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 303/ 311] blk.27.attn_norm.weight              - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 304/ 311] blk.27.attn_output.weight            - [  2048,   1024,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 305/ 311] blk.27.attn_q.weight                 - [  1024,   2048,      1,      1], type =   q8_0, converting to q4_K .. size =     2.12 MiB ->     1.12 MiB
[ 306/ 311] blk.27.attn_q_norm.weight            - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
[ 307/ 311] blk.27.attn_v.weight                 - [  1024,   1024,      1,      1], type =   q8_0, converting to q6_K .. size =     1.06 MiB ->     0.82 MiB
[ 308/ 311] blk.27.ffn_down.weight               - [  3072,   1024,      1,      1], type =   q8_0, converting to q6_K .. size =     3.19 MiB ->     2.46 MiB
[ 309/ 311] blk.27.ffn_gate.weight               - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
[ 310/ 311] blk.27.ffn_norm.weight               - [  1024,      1,      1,      1], type =    f32, size =    0.004 MiB
[ 311/ 311] blk.27.ffn_up.weight                 - [  1024,   3072,      1,      1], type =   q8_0, converting to q4_K .. size =     3.19 MiB ->     1.69 MiB
llama_model_quantize_impl: model size  =   761.80 MiB (8.50 BPW)
llama_model_quantize_impl: quant size  =   456.11 MiB (5.09 BPW)

llama_quantize: quantize time =  5188.91 ms
llama_quantize:    total time =  5188.91 ms
+ echo PASS
PASS
+ echo

+ /root/llama.me/build-ci-release/bin/llama-completion -no-cnv --model /root/llama.me/tmp/mnt/models/quantize/ggml-model-requant-merge.gguf -p 'I believe the meaning of life is' --n-predict 32
0.00.001.835 I llama_completion: llama backend init
0.00.001.841 I llama_completion: load the model and apply lora adapter, if any
0.00.001.849 I common_init_result: fitting params to device memory ...
0.00.001.850 I common_init_result: (for bugs during this step try to reproduce them with -fit off, or provide --verbose logs if the bug only occurs with -fit on)
0.00.176.375 I common_params_fit_impl: projected to use 5032 MiB of host memory vs. 77898 MiB of total host memory
0.00.257.352 W load: control-looking token: 128247 '</s>' was not control-type; this is probably a bug in the model. its type will be overridden
0.02.093.128 I common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
0.02.164.422 I llama_completion: llama threadpool init, n_threads = 8
0.02.164.449 I 
0.02.164.507 I system_info: n_threads = 8 (n_threads_batch = 8) / 16 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 | 
0.02.164.509 I 
0.02.164.595 I sampler seed: 1205069089
0.02.164.606 I sampler params: 
	repeat_last_n = 64, repeat_penalty = 1,000, frequency_penalty = 0,000, presence_penalty = 0,000
	dry_multiplier = 0,000, dry_base = 1,750, dry_allowed_length = 2, dry_penalty_last_n = -1
	top_k = 40, top_p = 0,950, min_p = 0,050, xtc_probability = 0,000, xtc_threshold = 0,100, typical_p = 1,000, top_n_sigma = -1,000, temp = 0,800
	mirostat = 0, mirostat_lr = 0,100, mirostat_ent = 5,000, adaptive_target = -1,000, adaptive_decay = 0,900
0.02.164.615 I sampler chain: logits -> ?penalties -> ?dry -> ?top-n-sigma -> top-k -> ?typical -> top-p -> min-p -> ?xtc -> temp-ext -> dist 
0.02.164.615 I generate: n_ctx = 40960, n_batch = 2048, n_predict = 32, n_keep = 0
0.02.164.615 I 
I believe the meaning of life is to live in the present moment and to create and share with others what can be done in the present. How do I get from today to tomorrow in the context

0.02.432.656 I common_perf_print:    sampling time =       5,04 ms
0.02.432.658 I common_perf_print:    samplers time =       1,64 ms /    39 tokens
0.02.432.664 I common_perf_print:        load time =    1949,26 ms
0.02.432.665 I common_perf_print: prompt eval time =      14,98 ms /     7 tokens (    2,14 ms per token,   467,13 tokens per second)
0.02.432.666 I common_perf_print:        eval time =     247,27 ms /    31 runs   (    7,98 ms per token,   125,37 tokens per second)
0.02.432.666 I common_perf_print:       total time =     268,25 ms /    38 tokens
0.02.432.666 I common_perf_print: unaccounted time =       0,94 ms /   0,4 %      (total - sampling - prompt eval - eval) / (total)
0.02.432.667 I common_perf_print:    graphs reused =         30
+ echo PASS
PASS
+ echo

+ rm -f /root/llama.me/tmp/mnt/models/quantize/ggml-model-split-00001-of-00012.gguf /root/llama.me/tmp/mnt/models/quantize/ggml-model-split-00002-of-00012.gguf /root/llama.me/tmp/mnt/models/quantize/ggml-model-split-00003-of-00012.gguf /root/llama.me/tmp/mnt/models/quantize/ggml-model-split-00004-of-00012.gguf /root/llama.me/tmp/mnt/models/quantize/ggml-model-split-00005-of-00012.gguf /root/llama.me/tmp/mnt/models/quantize/ggml-model-split-00006-of-00012.gguf /root/llama.me/tmp/mnt/models/quantize/ggml-model-split-00007-of-00012.gguf /root/llama.me/tmp/mnt/models/quantize/ggml-model-split-00008-of-00012.gguf /root/llama.me/tmp/mnt/models/quantize/ggml-model-split-00009-of-00012.gguf /root/llama.me/tmp/mnt/models/quantize/ggml-model-split-00010-of-00012.gguf /root/llama.me/tmp/mnt/models/quantize/ggml-model-split-00011-of-00012.gguf /root/llama.me/tmp/mnt/models/quantize/ggml-model-split-00012-of-00012.gguf /root/llama.me/tmp/mnt/models/quantize/ggml-model-requant-00001-of-00012.gguf /root/llama.me/tmp/mnt/models/quantize/ggml-model-requant-00002-of-00012.gguf /root/llama.me/tmp/mnt/models/quantize/ggml-model-requant-00003-of-00012.gguf /root/llama.me/tmp/mnt/models/quantize/ggml-model-requant-00004-of-00012.gguf /root/llama.me/tmp/mnt/models/quantize/ggml-model-requant-00005-of-00012.gguf /root/llama.me/tmp/mnt/models/quantize/ggml-model-requant-00006-of-00012.gguf /root/llama.me/tmp/mnt/models/quantize/ggml-model-requant-00007-of-00012.gguf /root/llama.me/tmp/mnt/models/quantize/ggml-model-requant-00008-of-00012.gguf /root/llama.me/tmp/mnt/models/quantize/ggml-model-requant-00009-of-00012.gguf /root/llama.me/tmp/mnt/models/quantize/ggml-model-requant-00010-of-00012.gguf /root/llama.me/tmp/mnt/models/quantize/ggml-model-requant-00011-of-00012.gguf /root/llama.me/tmp/mnt/models/quantize/ggml-model-requant-00012-of-00012.gguf /root/llama.me/tmp/mnt/models/quantize/ggml-model-requant-merge.gguf

real	0m16,949s
user	2m2,505s
sys	0m6,519s

qwen3_0_6b

Qwen3 0.6B:

  • status: 1

  • perplexity:

  • imatrix:


  • f16:
+ ./bin/llama-completion -no-cnv --model ../models-mnt/qwen3/0.6B/ggml-model-f16.gguf -ngl 99 -c 1024 -s 1234 -n 64 --ignore-eos -p 'I believe the meaning of life is'
warning: no usable GPU found, --gpu-layers option will be ignored
warning: one possible reason is that llama.cpp was compiled without GPU support
warning: consult docs/build.md for compilation instructions
0.00.001.909 I llama_completion: llama backend init
0.00.001.913 I llama_completion: load the model and apply lora adapter, if any
0.00.001.920 I common_init_result: fitting params to device memory ...
0.00.001.921 I common_init_result: (for bugs during this step try to reproduce them with -fit off, or provide --verbose logs if the bug only occurs with -fit on)
0.00.185.993 I common_params_fit_impl: projected to use 1558 MiB of host memory vs. 77898 MiB of total host memory
0.00.273.239 W load: control-looking token: 128247 '</s>' was not control-type; this is probably a bug in the model. its type will be overridden
0.00.401.484 W llama_context: n_ctx_seq (1024) < n_ctx_train (32768) -- the full capacity of the model will not be utilized
0.00.444.381 I common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
0.00.474.552 I llama_completion: llama threadpool init, n_threads = 8
0.00.474.573 I 
0.00.474.643 I system_info: n_threads = 8 (n_threads_batch = 8) / 16 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 | 
0.00.474.644 I 
0.00.474.716 I sampler seed: 1234
0.00.474.725 I sampler params: 
	repeat_last_n = 64, repeat_penalty = 1,000, frequency_penalty = 0,000, presence_penalty = 0,000
	dry_multiplier = 0,000, dry_base = 1,750, dry_allowed_length = 2, dry_penalty_last_n = -1
	top_k = 40, top_p = 0,950, min_p = 0,050, xtc_probability = 0,000, xtc_threshold = 0,100, typical_p = 1,000, top_n_sigma = -1,000, temp = 0,800
	mirostat = 0, mirostat_lr = 0,100, mirostat_ent = 5,000, adaptive_target = -1,000, adaptive_decay = 0,900
0.00.474.731 I sampler chain: logits -> logit-bias -> ?penalties -> ?dry -> ?top-n-sigma -> top-k -> ?typical -> top-p -> min-p -> ?xtc -> temp-ext -> dist 
0.00.474.732 I generate: n_ctx = 1024, n_batch = 2048, n_predict = 64, n_keep = 0
0.00.474.732 I 
I believe the meaning of life is to have fun, so I have made a list of activities that I love doing. I have a lot of them but some of them are just more fun than others so I am going to list them one by one. One of my favourite places to go is to a park or a beach. I love going to these

0.01.884.667 I common_perf_print:    sampling time =       8,24 ms
0.01.884.669 I common_perf_print:    samplers time =       3,32 ms /    71 tokens
0.01.884.675 I common_perf_print:        load time =     249,29 ms
0.01.884.676 I common_perf_print: prompt eval time =      24,02 ms /     7 tokens (    3,43 ms per token,   291,48 tokens per second)
0.01.884.677 I common_perf_print:        eval time =    1376,15 ms /    63 runs   (   21,84 ms per token,    45,78 tokens per second)
0.01.884.678 I common_perf_print:       total time =    1410,12 ms /    70 tokens
0.01.884.678 I common_perf_print: unaccounted time =       1,73 ms /   0,1 %      (total - sampling - prompt eval - eval) / (total)
0.01.884.678 I common_perf_print:    graphs reused =         62

real	0m1,963s
user	0m11,850s
sys	0m0,218s
+ ./bin/llama-perplexity --model ../models-mnt/qwen3/0.6B/ggml-model-f16.gguf -f ../models-mnt/wikitext/wikitext-2-raw/wiki.test.raw -ngl 99 -c 1024 -b 512 --chunks 2
error while handling argument "-f": error: failed to open file '../models-mnt/wikitext/wikitext-2-raw/wiki.test.raw'


usage:
-f,    --file FNAME                     a file containing the prompt (default: none)


to show complete usage, run with -h

real	0m0,005s
user	0m0,003s
sys	0m0,002s
  • bf16:
+ ./bin/llama-completion -no-cnv --model ../models-mnt/qwen3/0.6B/ggml-model-bf16.gguf -ngl 99 -c 1024 -s 1234 -n 64 --ignore-eos -p 'I believe the meaning of life is'
warning: no usable GPU found, --gpu-layers option will be ignored
warning: one possible reason is that llama.cpp was compiled without GPU support
warning: consult docs/build.md for compilation instructions
0.00.001.802 I llama_completion: llama backend init
0.00.001.806 I llama_completion: load the model and apply lora adapter, if any
0.00.001.814 I common_init_result: fitting params to device memory ...
0.00.001.814 I common_init_result: (for bugs during this step try to reproduce them with -fit off, or provide --verbose logs if the bug only occurs with -fit on)
0.00.179.409 I common_params_fit_impl: projected to use 1558 MiB of host memory vs. 77898 MiB of total host memory
0.00.261.991 W load: control-looking token: 128247 '</s>' was not control-type; this is probably a bug in the model. its type will be overridden
0.00.379.053 W llama_context: n_ctx_seq (1024) < n_ctx_train (32768) -- the full capacity of the model will not be utilized
0.00.423.666 I common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
0.00.453.866 I llama_completion: llama threadpool init, n_threads = 8
0.00.453.900 I 
0.00.454.001 I system_info: n_threads = 8 (n_threads_batch = 8) / 16 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 | 
0.00.454.004 I 
0.00.454.125 I sampler seed: 1234
0.00.454.138 I sampler params: 
	repeat_last_n = 64, repeat_penalty = 1,000, frequency_penalty = 0,000, presence_penalty = 0,000
	dry_multiplier = 0,000, dry_base = 1,750, dry_allowed_length = 2, dry_penalty_last_n = -1
	top_k = 40, top_p = 0,950, min_p = 0,050, xtc_probability = 0,000, xtc_threshold = 0,100, typical_p = 1,000, top_n_sigma = -1,000, temp = 0,800
	mirostat = 0, mirostat_lr = 0,100, mirostat_ent = 5,000, adaptive_target = -1,000, adaptive_decay = 0,900
0.00.454.148 I sampler chain: logits -> logit-bias -> ?penalties -> ?dry -> ?top-n-sigma -> top-k -> ?typical -> top-p -> min-p -> ?xtc -> temp-ext -> dist 
0.00.454.151 I generate: n_ctx = 1024, n_batch = 2048, n_predict = 64, n_keep = 0
0.00.454.151 I 
I believe the meaning of life is to give, not to receive. Do you agree with this view?

Yes, I agree with this view. The concept of giving as the foundation of life reflects a deeper spiritual principle. It suggests that the purpose of existence is to contribute to the well-being of others and society, rather than seeking personal gain. This perspective

0.01.887.219 I common_perf_print:    sampling time =       8,23 ms
0.01.887.221 I common_perf_print:    samplers time =       3,12 ms /    71 tokens
0.01.887.227 I common_perf_print:        load time =     235,29 ms
0.01.887.228 I common_perf_print: prompt eval time =      24,25 ms /     7 tokens (    3,46 ms per token,   288,60 tokens per second)
0.01.887.229 I common_perf_print:        eval time =    1398,91 ms /    63 runs   (   22,20 ms per token,    45,04 tokens per second)
0.01.887.229 I common_perf_print:       total time =    1433,36 ms /    70 tokens
0.01.887.229 I common_perf_print: unaccounted time =       1,97 ms /   0,1 %      (total - sampling - prompt eval - eval) / (total)
0.01.887.229 I common_perf_print:    graphs reused =         62

real	0m1,962s
user	0m12,018s
sys	0m0,209s
  • q8_0:
+ ./bin/llama-completion -no-cnv --model ../models-mnt/qwen3/0.6B/ggml-model-q8_0.gguf -ngl 99 -c 1024 -s 1234 -n 64 --ignore-eos -p 'I believe the meaning of life is'
warning: no usable GPU found, --gpu-layers option will be ignored
warning: one possible reason is that llama.cpp was compiled without GPU support
warning: consult docs/build.md for compilation instructions
0.00.001.685 I llama_completion: llama backend init
0.00.001.689 I llama_completion: load the model and apply lora adapter, if any
0.00.001.696 I common_init_result: fitting params to device memory ...
0.00.001.696 I common_init_result: (for bugs during this step try to reproduce them with -fit off, or provide --verbose logs if the bug only occurs with -fit on)
0.00.181.291 I common_params_fit_impl: projected to use 1025 MiB of host memory vs. 77898 MiB of total host memory
0.00.262.973 W load: control-looking token: 128247 '</s>' was not control-type; this is probably a bug in the model. its type will be overridden
0.00.346.508 W llama_context: n_ctx_seq (1024) < n_ctx_train (32768) -- the full capacity of the model will not be utilized
0.00.391.419 I common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
0.00.414.329 I llama_completion: llama threadpool init, n_threads = 8
0.00.414.349 I 
0.00.414.416 I system_info: n_threads = 8 (n_threads_batch = 8) / 16 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 | 
0.00.414.417 I 
0.00.414.491 I sampler seed: 1234
0.00.414.498 I sampler params: 
	repeat_last_n = 64, repeat_penalty = 1,000, frequency_penalty = 0,000, presence_penalty = 0,000
	dry_multiplier = 0,000, dry_base = 1,750, dry_allowed_length = 2, dry_penalty_last_n = -1
	top_k = 40, top_p = 0,950, min_p = 0,050, xtc_probability = 0,000, xtc_threshold = 0,100, typical_p = 1,000, top_n_sigma = -1,000, temp = 0,800
	mirostat = 0, mirostat_lr = 0,100, mirostat_ent = 5,000, adaptive_target = -1,000, adaptive_decay = 0,900
0.00.414.505 I sampler chain: logits -> logit-bias -> ?penalties -> ?dry -> ?top-n-sigma -> top-k -> ?typical -> top-p -> min-p -> ?xtc -> temp-ext -> dist 
0.00.414.505 I generate: n_ctx = 1024, n_batch = 2048, n_predict = 64, n_keep = 0
0.00.414.506 I 
I believe the meaning of life is to give back. That's why I'm a volunteer. I am a volunteer because I believe that life is important to give back. I work in a nursing home to help people. I also help with the medical staff at a clinic. I am also a volunteer for the Red Cross. I volunteer because I think giving

0.01.218.670 I common_perf_print:    sampling time =       9,11 ms
0.01.218.672 I common_perf_print:    samplers time =       3,60 ms /    71 tokens
0.01.218.677 I common_perf_print:        load time =     194,18 ms
0.01.218.681 I common_perf_print: prompt eval time =      16,42 ms /     7 tokens (    2,35 ms per token,   426,28 tokens per second)
0.01.218.682 I common_perf_print:        eval time =     777,29 ms /    63 runs   (   12,34 ms per token,    81,05 tokens per second)
0.01.218.682 I common_perf_print:       total time =     804,35 ms /    70 tokens
0.01.218.686 I common_perf_print: unaccounted time =       1,52 ms /   0,2 %      (total - sampling - prompt eval - eval) / (total)
0.01.218.686 I common_perf_print:    graphs reused =         62

real	0m1,277s
user	0m6,927s
sys	0m0,166s
  • q4_0:
+ ./bin/llama-completion -no-cnv --model ../models-mnt/qwen3/0.6B/ggml-model-q4_0.gguf -ngl 99 -c 1024 -s 1234 -n 64 --ignore-eos -p 'I believe the meaning of life is'
warning: no usable GPU found, --gpu-layers option will be ignored
warning: one possible reason is that llama.cpp was compiled without GPU support
warning: consult docs/build.md for compilation instructions
0.00.001.703 I llama_completion: llama backend init
0.00.001.707 I llama_completion: load the model and apply lora adapter, if any
0.00.001.715 I common_init_result: fitting params to device memory ...
0.00.001.715 I common_init_result: (for bugs during this step try to reproduce them with -fit off, or provide --verbose logs if the bug only occurs with -fit on)
0.00.181.396 I common_params_fit_impl: projected to use 543 MiB of host memory vs. 77898 MiB of total host memory
0.00.265.419 W load: control-looking token: 128247 '</s>' was not control-type; this is probably a bug in the model. its type will be overridden
0.00.461.489 W llama_context: n_ctx_seq (1024) < n_ctx_train (32768) -- the full capacity of the model will not be utilized
0.00.506.112 I common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
0.00.520.832 I llama_completion: llama threadpool init, n_threads = 8
0.00.520.852 I 
0.00.520.922 I system_info: n_threads = 8 (n_threads_batch = 8) / 16 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 | 
0.00.520.923 I 
0.00.520.998 I sampler seed: 1234
0.00.521.007 I sampler params: 
	repeat_last_n = 64, repeat_penalty = 1,000, frequency_penalty = 0,000, presence_penalty = 0,000
	dry_multiplier = 0,000, dry_base = 1,750, dry_allowed_length = 2, dry_penalty_last_n = -1
	top_k = 40, top_p = 0,950, min_p = 0,050, xtc_probability = 0,000, xtc_threshold = 0,100, typical_p = 1,000, top_n_sigma = -1,000, temp = 0,800
	mirostat = 0, mirostat_lr = 0,100, mirostat_ent = 5,000, adaptive_target = -1,000, adaptive_decay = 0,900
0.00.521.013 I sampler chain: logits -> logit-bias -> ?penalties -> ?dry -> ?top-n-sigma -> top-k -> ?typical -> top-p -> min-p -> ?xtc -> temp-ext -> dist 
0.00.521.013 I generate: n_ctx = 1024, n_batch = 2048, n_predict = 64, n_keep = 0
0.00.521.013 I 
I believe the meaning of life is to make the most out of my life, and to be happy. This is what I have been doing. I'm not sure I have reached my full potential, and I'm not sure if I have given myself enough freedom to grow spiritually.

What's the point of life?

I'm not sure I have reached my

0.01.026.627 I common_perf_print:    sampling time =       9,60 ms
0.01.026.628 I common_perf_print:    samplers time =       3,90 ms /    71 tokens
0.01.026.633 I common_perf_print:        load time =     299,50 ms
0.01.026.635 I common_perf_print: prompt eval time =      14,61 ms /     7 tokens (    2,09 ms per token,   479,26 tokens per second)
0.01.026.636 I common_perf_print:        eval time =     480,01 ms /    63 runs   (    7,62 ms per token,   131,25 tokens per second)
0.01.026.636 I common_perf_print:       total time =     505,80 ms /    70 tokens
0.01.026.637 I common_perf_print: unaccounted time =       1,59 ms /   0,3 %      (total - sampling - prompt eval - eval) / (total)
0.01.026.637 I common_perf_print:    graphs reused =         62

real	0m1,101s
user	0m4,548s
sys	0m0,221s
  • q4_1:
+ ./bin/llama-completion -no-cnv --model ../models-mnt/qwen3/0.6B/ggml-model-q4_1.gguf -ngl 99 -c 1024 -s 1234 -n 64 --ignore-eos -p 'I believe the meaning of life is'
warning: no usable GPU found, --gpu-layers option will be ignored
warning: one possible reason is that llama.cpp was compiled without GPU support
warning: consult docs/build.md for compilation instructions
0.00.001.728 I llama_completion: llama backend init
0.00.001.740 I llama_completion: load the model and apply lora adapter, if any
0.00.001.748 I common_init_result: fitting params to device memory ...
0.00.001.748 I common_init_result: (for bugs during this step try to reproduce them with -fit off, or provide --verbose logs if the bug only occurs with -fit on)
0.00.181.720 I common_params_fit_impl: projected to use 806 MiB of host memory vs. 77898 MiB of total host memory
0.00.262.978 W load: control-looking token: 128247 '</s>' was not control-type; this is probably a bug in the model. its type will be overridden
0.00.333.540 W llama_context: n_ctx_seq (1024) < n_ctx_train (32768) -- the full capacity of the model will not be utilized
0.00.378.316 I common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
0.00.394.368 I llama_completion: llama threadpool init, n_threads = 8
0.00.394.389 I 
0.00.394.458 I system_info: n_threads = 8 (n_threads_batch = 8) / 16 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 | 
0.00.394.459 I 
0.00.394.535 I sampler seed: 1234
0.00.394.544 I sampler params: 
	repeat_last_n = 64, repeat_penalty = 1,000, frequency_penalty = 0,000, presence_penalty = 0,000
	dry_multiplier = 0,000, dry_base = 1,750, dry_allowed_length = 2, dry_penalty_last_n = -1
	top_k = 40, top_p = 0,950, min_p = 0,050, xtc_probability = 0,000, xtc_threshold = 0,100, typical_p = 1,000, top_n_sigma = -1,000, temp = 0,800
	mirostat = 0, mirostat_lr = 0,100, mirostat_ent = 5,000, adaptive_target = -1,000, adaptive_decay = 0,900
0.00.394.550 I sampler chain: logits -> logit-bias -> ?penalties -> ?dry -> ?top-n-sigma -> top-k -> ?typical -> top-p -> min-p -> ?xtc -> temp-ext -> dist 
0.00.394.551 I generate: n_ctx = 1024, n_batch = 2048, n_predict = 64, n_keep = 0
0.00.394.551 I 
I believe the meaning of life is to find your own voice. I will never be able to hear my own voice. I will never be able to listen to a song. I will never be able to see my own face. I will never be able to share my own thoughts. I will never be able to share my own voice. I will never

0.00.957.264 I common_perf_print:    sampling time =       9,97 ms
0.00.957.265 I common_perf_print:    samplers time =       4,08 ms /    71 tokens
0.00.957.270 I common_perf_print:        load time =     173,96 ms
0.00.957.271 I common_perf_print: prompt eval time =      20,90 ms /     7 tokens (    2,99 ms per token,   334,94 tokens per second)
0.00.957.271 I common_perf_print:        eval time =     530,36 ms /    63 runs   (    8,42 ms per token,   118,79 tokens per second)
0.00.957.272 I common_perf_print:       total time =     562,90 ms /    70 tokens
0.00.957.272 I common_perf_print: unaccounted time =       1,67 ms /   0,3 %      (total - sampling - prompt eval - eval) / (total)
0.00.957.272 I common_perf_print:    graphs reused =         62

real	0m1,008s
user	0m4,952s
sys	0m0,134s
  • q5_0:
+ ./bin/llama-completion -no-cnv --model ../models-mnt/qwen3/0.6B/ggml-model-q5_0.gguf -ngl 99 -c 1024 -s 1234 -n 64 --ignore-eos -p 'I believe the meaning of life is'
warning: no usable GPU found, --gpu-layers option will be ignored
warning: one possible reason is that llama.cpp was compiled without GPU support
warning: consult docs/build.md for compilation instructions
0.00.001.705 I llama_completion: llama backend init
0.00.001.709 I llama_completion: load the model and apply lora adapter, if any
0.00.001.716 I common_init_result: fitting params to device memory ...
0.00.001.716 I common_init_result: (for bugs during this step try to reproduce them with -fit off, or provide --verbose logs if the bug only occurs with -fit on)
0.00.181.683 I common_params_fit_impl: projected to use 832 MiB of host memory vs. 77898 MiB of total host memory
0.00.270.458 W load: control-looking token: 128247 '</s>' was not control-type; this is probably a bug in the model. its type will be overridden
0.00.345.237 W llama_context: n_ctx_seq (1024) < n_ctx_train (32768) -- the full capacity of the model will not be utilized
0.00.387.599 I common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
0.00.404.291 I llama_completion: llama threadpool init, n_threads = 8
0.00.404.310 I 
0.00.404.378 I system_info: n_threads = 8 (n_threads_batch = 8) / 16 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 | 
0.00.404.380 I 
0.00.404.452 I sampler seed: 1234
0.00.404.460 I sampler params: 
	repeat_last_n = 64, repeat_penalty = 1,000, frequency_penalty = 0,000, presence_penalty = 0,000
	dry_multiplier = 0,000, dry_base = 1,750, dry_allowed_length = 2, dry_penalty_last_n = -1
	top_k = 40, top_p = 0,950, min_p = 0,050, xtc_probability = 0,000, xtc_threshold = 0,100, typical_p = 1,000, top_n_sigma = -1,000, temp = 0,800
	mirostat = 0, mirostat_lr = 0,100, mirostat_ent = 5,000, adaptive_target = -1,000, adaptive_decay = 0,900
0.00.404.466 I sampler chain: logits -> logit-bias -> ?penalties -> ?dry -> ?top-n-sigma -> top-k -> ?typical -> top-p -> min-p -> ?xtc -> temp-ext -> dist 
0.00.404.468 I generate: n_ctx = 1024, n_batch = 2048, n_predict = 64, n_keep = 0
0.00.404.468 I 
I believe the meaning of life is to give a chance for others to feel the joy of life. Is this statement correct?

Yes, your statement is correct. Many people believe that the purpose of life is to inspire and uplift others, allowing them to experience happiness and fulfillment. This perspective emphasizes the importance of compassion, empathy, and the impact we have on

0.00.992.428 I common_perf_print:    sampling time =       8,09 ms
0.00.992.431 I common_perf_print:    samplers time =       3,21 ms /    71 tokens
0.00.992.434 I common_perf_print:        load time =     183,44 ms
0.00.992.435 I common_perf_print: prompt eval time =      14,92 ms /     7 tokens (    2,13 ms per token,   469,04 tokens per second)
0.00.992.436 I common_perf_print:        eval time =     563,63 ms /    63 runs   (    8,95 ms per token,   111,77 tokens per second)
0.00.992.436 I common_perf_print:       total time =     588,14 ms /    70 tokens
0.00.992.437 I common_perf_print: unaccounted time =       1,50 ms /   0,3 %      (total - sampling - prompt eval - eval) / (total)
0.00.992.437 I common_perf_print:    graphs reused =         62

real	0m1,044s
user	0m5,176s
sys	0m0,124s
  • q5_1:
+ ./bin/llama-completion -no-cnv --model ../models-mnt/qwen3/0.6B/ggml-model-q5_1.gguf -ngl 99 -c 1024 -s 1234 -n 64 --ignore-eos -p 'I believe the meaning of life is'
warning: no usable GPU found, --gpu-layers option will be ignored
warning: one possible reason is that llama.cpp was compiled without GPU support
warning: consult docs/build.md for compilation instructions
0.00.001.731 I llama_completion: llama backend init
0.00.001.736 I llama_completion: load the model and apply lora adapter, if any
0.00.001.744 I common_init_result: fitting params to device memory ...
0.00.001.744 I common_init_result: (for bugs during this step try to reproduce them with -fit off, or provide --verbose logs if the bug only occurs with -fit on)
0.00.184.383 I common_params_fit_impl: projected to use 858 MiB of host memory vs. 77898 MiB of total host memory
0.00.266.357 W load: control-looking token: 128247 '</s>' was not control-type; this is probably a bug in the model. its type will be overridden
0.00.339.606 W llama_context: n_ctx_seq (1024) < n_ctx_train (32768) -- the full capacity of the model will not be utilized
0.00.384.268 I common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
0.00.401.346 I llama_completion: llama threadpool init, n_threads = 8
0.00.401.365 I 
0.00.401.425 I system_info: n_threads = 8 (n_threads_batch = 8) / 16 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 | 
0.00.401.427 I 
0.00.401.495 I sampler seed: 1234
0.00.401.502 I sampler params: 
	repeat_last_n = 64, repeat_penalty = 1,000, frequency_penalty = 0,000, presence_penalty = 0,000
	dry_multiplier = 0,000, dry_base = 1,750, dry_allowed_length = 2, dry_penalty_last_n = -1
	top_k = 40, top_p = 0,950, min_p = 0,050, xtc_probability = 0,000, xtc_threshold = 0,100, typical_p = 1,000, top_n_sigma = -1,000, temp = 0,800
	mirostat = 0, mirostat_lr = 0,100, mirostat_ent = 5,000, adaptive_target = -1,000, adaptive_decay = 0,900
0.00.401.507 I sampler chain: logits -> logit-bias -> ?penalties -> ?dry -> ?top-n-sigma -> top-k -> ?typical -> top-p -> min-p -> ?xtc -> temp-ext -> dist 
0.00.401.510 I generate: n_ctx = 1024, n_batch = 2048, n_predict = 64, n_keep = 0
0.00.401.510 I 
I believe the meaning of life is to give and to be given. I have been given a lot of responsibility, and I have been given the ability to use that responsibility to help others. I have always been very passionate about helping others. I have seen a positive effect from my actions and I believe that this is the meaning of life. I have never

0.01.026.156 I common_perf_print:    sampling time =       9,28 ms
0.01.026.156 I common_perf_print:    samplers time =       3,70 ms /    71 tokens
0.01.026.160 I common_perf_print:        load time =     178,23 ms
0.01.026.161 I common_perf_print: prompt eval time =      24,06 ms /     7 tokens (    3,44 ms per token,   290,96 tokens per second)
0.01.026.161 I common_perf_print:        eval time =     590,06 ms /    63 runs   (    9,37 ms per token,   106,77 tokens per second)
0.01.026.174 I common_perf_print:       total time =     624,81 ms /    70 tokens
0.01.026.175 I common_perf_print: unaccounted time =       1,41 ms /   0,2 %      (total - sampling - prompt eval - eval) / (total)
0.01.026.176 I common_perf_print:    graphs reused =         62

real	0m1,080s
user	0m5,472s
sys	0m0,126s
  • q2_k:
+ ./bin/llama-completion -no-cnv --model ../models-mnt/qwen3/0.6B/ggml-model-q2_k.gguf -ngl 99 -c 1024 -s 1234 -n 64 --ignore-eos -p 'I believe the meaning of life is'
warning: no usable GPU found, --gpu-layers option will be ignored
warning: one possible reason is that llama.cpp was compiled without GPU support
warning: consult docs/build.md for compilation instructions
0.00.002.477 I llama_completion: llama backend init
0.00.002.486 I llama_completion: load the model and apply lora adapter, if any
0.00.002.501 I common_init_result: fitting params to device memory ...
0.00.002.502 I common_init_result: (for bugs during this step try to reproduce them with -fit off, or provide --verbose logs if the bug only occurs with -fit on)
0.00.186.936 I common_params_fit_impl: projected to use 615 MiB of host memory vs. 77898 MiB of total host memory
0.00.276.361 W load: control-looking token: 128247 '</s>' was not control-type; this is probably a bug in the model. its type will be overridden
0.00.393.253 W llama_context: n_ctx_seq (1024) < n_ctx_train (32768) -- the full capacity of the model will not be utilized
0.00.438.008 I common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
0.00.450.969 I llama_completion: llama threadpool init, n_threads = 8
0.00.451.004 I 
0.00.451.073 I system_info: n_threads = 8 (n_threads_batch = 8) / 16 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 | 
0.00.451.075 I 
0.00.451.156 I sampler seed: 1234
0.00.451.166 I sampler params: 
	repeat_last_n = 64, repeat_penalty = 1,000, frequency_penalty = 0,000, presence_penalty = 0,000
	dry_multiplier = 0,000, dry_base = 1,750, dry_allowed_length = 2, dry_penalty_last_n = -1
	top_k = 40, top_p = 0,950, min_p = 0,050, xtc_probability = 0,000, xtc_threshold = 0,100, typical_p = 1,000, top_n_sigma = -1,000, temp = 0,800
	mirostat = 0, mirostat_lr = 0,100, mirostat_ent = 5,000, adaptive_target = -1,000, adaptive_decay = 0,900
0.00.451.174 I sampler chain: logits -> logit-bias -> ?penalties -> ?dry -> ?top-n-sigma -> top-k -> ?typical -> top-p -> min-p -> ?xtc -> temp-ext -> dist 
0.00.451.177 I generate: n_ctx = 1024, n_batch = 2048, n_predict = 64, n_keep = 0
0.00.451.177 I 
I believe the meaning of life is not of the soul is of the soul
A. is of the body
B. is of the brain
C. is of the brain
C. is of the mind
C. is of the mind
C. is of the mind
C. is of the soul
C. is of the soul

0.00.879.249 I common_perf_print:    sampling time =       9,28 ms
0.00.879.250 I common_perf_print:    samplers time =       3,73 ms /    71 tokens
0.00.879.254 I common_perf_print:        load time =     223,40 ms
0.00.879.255 I common_perf_print: prompt eval time =      15,96 ms /     7 tokens (    2,28 ms per token,   438,62 tokens per second)
0.00.879.256 I common_perf_print:        eval time =     401,44 ms /    63 runs   (    6,37 ms per token,   156,94 tokens per second)
0.00.879.256 I common_perf_print:       total time =     428,29 ms /    70 tokens
0.00.879.257 I common_perf_print: unaccounted time =       1,61 ms /   0,4 %      (total - sampling - prompt eval - eval) / (total)
0.00.879.257 I common_perf_print:    graphs reused =         62

real	0m0,929s
user	0m3,878s
sys	0m0,163s
  • q3_k:
+ ./bin/llama-completion -no-cnv --model ../models-mnt/qwen3/0.6B/ggml-model-q3_k.gguf -ngl 99 -c 1024 -s 1234 -n 64 --ignore-eos -p 'I believe the meaning of life is'
warning: no usable GPU found, --gpu-layers option will be ignored
warning: one possible reason is that llama.cpp was compiled without GPU support
warning: consult docs/build.md for compilation instructions
0.00.001.753 I llama_completion: llama backend init
0.00.001.757 I llama_completion: load the model and apply lora adapter, if any
0.00.001.764 I common_init_result: fitting params to device memory ...
0.00.001.765 I common_init_result: (for bugs during this step try to reproduce them with -fit off, or provide --verbose logs if the bug only occurs with -fit on)
0.00.204.313 I common_params_fit_impl: projected to use 655 MiB of host memory vs. 77898 MiB of total host memory
0.00.286.487 W load: control-looking token: 128247 '</s>' was not control-type; this is probably a bug in the model. its type will be overridden
0.00.437.765 W llama_context: n_ctx_seq (1024) < n_ctx_train (32768) -- the full capacity of the model will not be utilized
0.00.483.458 I common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
0.00.498.255 I llama_completion: llama threadpool init, n_threads = 8
0.00.498.277 I 
0.00.498.349 I system_info: n_threads = 8 (n_threads_batch = 8) / 16 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 | 
0.00.498.350 I 
0.00.498.429 I sampler seed: 1234
0.00.498.438 I sampler params: 
	repeat_last_n = 64, repeat_penalty = 1,000, frequency_penalty = 0,000, presence_penalty = 0,000
	dry_multiplier = 0,000, dry_base = 1,750, dry_allowed_length = 2, dry_penalty_last_n = -1
	top_k = 40, top_p = 0,950, min_p = 0,050, xtc_probability = 0,000, xtc_threshold = 0,100, typical_p = 1,000, top_n_sigma = -1,000, temp = 0,800
	mirostat = 0, mirostat_lr = 0,100, mirostat_ent = 5,000, adaptive_target = -1,000, adaptive_decay = 0,900
0.00.498.447 I sampler chain: logits -> logit-bias -> ?penalties -> ?dry -> ?top-n-sigma -> top-k -> ?typical -> top-p -> min-p -> ?xtc -> temp-ext -> dist 
0.00.498.448 I generate: n_ctx = 1024, n_batch = 2048, n_predict = 64, n_keep = 0
0.00.498.448 I 
I believe the meaning of life is to serve God, but I don’t know how to live it. I don’t know how to live it, so how can I serve God? And I don’t know if I am to serve God. What can I serve God, and where will I serve God? Also, I know that I am to serve

0.00.982.197 I common_perf_print:    sampling time =       9,62 ms
0.00.982.198 I common_perf_print:    samplers time =       3,99 ms /    71 tokens
0.00.982.201 I common_perf_print:        load time =     254,85 ms
0.00.982.202 I common_perf_print: prompt eval time =      17,17 ms /     7 tokens (    2,45 ms per token,   407,74 tokens per second)
0.00.982.203 I common_perf_print:        eval time =     455,75 ms /    63 runs   (    7,23 ms per token,   138,23 tokens per second)
0.00.982.203 I common_perf_print:       total time =     483,95 ms /    70 tokens
0.00.982.204 I common_perf_print: unaccounted time =       1,41 ms /   0,3 %      (total - sampling - prompt eval - eval) / (total)
0.00.982.204 I common_perf_print:    graphs reused =         62

real	0m1,038s
user	0m4,392s
sys	0m0,161s
  • q4_k:
+ ./bin/llama-completion -no-cnv --model ../models-mnt/qwen3/0.6B/ggml-model-q4_k.gguf -ngl 99 -c 1024 -s 1234 -n 64 --ignore-eos -p 'I believe the meaning of life is'
warning: no usable GPU found, --gpu-layers option will be ignored
warning: one possible reason is that llama.cpp was compiled without GPU support
warning: consult docs/build.md for compilation instructions
0.00.001.878 I llama_completion: llama backend init
0.00.001.888 I llama_completion: load the model and apply lora adapter, if any
0.00.001.901 I common_init_result: fitting params to device memory ...
0.00.001.901 I common_init_result: (for bugs during this step try to reproduce them with -fit off, or provide --verbose logs if the bug only occurs with -fit on)
0.00.190.336 I common_params_fit_impl: projected to use 589 MiB of host memory vs. 77898 MiB of total host memory
0.00.275.111 W load: control-looking token: 128247 '</s>' was not control-type; this is probably a bug in the model. its type will be overridden
0.00.533.423 W llama_context: n_ctx_seq (1024) < n_ctx_train (32768) -- the full capacity of the model will not be utilized
0.00.575.479 I common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
0.00.590.624 I llama_completion: llama threadpool init, n_threads = 8
0.00.590.645 I 
0.00.590.709 I system_info: n_threads = 8 (n_threads_batch = 8) / 16 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 | 
0.00.590.710 I 
0.00.590.778 I sampler seed: 1234
0.00.590.786 I sampler params: 
	repeat_last_n = 64, repeat_penalty = 1,000, frequency_penalty = 0,000, presence_penalty = 0,000
	dry_multiplier = 0,000, dry_base = 1,750, dry_allowed_length = 2, dry_penalty_last_n = -1
	top_k = 40, top_p = 0,950, min_p = 0,050, xtc_probability = 0,000, xtc_threshold = 0,100, typical_p = 1,000, top_n_sigma = -1,000, temp = 0,800
	mirostat = 0, mirostat_lr = 0,100, mirostat_ent = 5,000, adaptive_target = -1,000, adaptive_decay = 0,900
0.00.590.792 I sampler chain: logits -> logit-bias -> ?penalties -> ?dry -> ?top-n-sigma -> top-k -> ?typical -> top-p -> min-p -> ?xtc -> temp-ext -> dist 
0.00.590.793 I generate: n_ctx = 1024, n_batch = 2048, n_predict = 64, n_keep = 0
0.00.590.793 I 
I believe the meaning of life is to do what is good and right and to be a good person. I believe that the only way to be good and right is to do what is good and right. I believe that by doing what is good and right, I will become a good and right person. The world would be a better place if I were

0.01.116.936 I common_perf_print:    sampling time =       9,01 ms
0.01.116.936 I common_perf_print:    samplers time =       3,62 ms /    71 tokens
0.01.116.941 I common_perf_print:        load time =     360,91 ms
0.01.116.942 I common_perf_print: prompt eval time =      13,12 ms /     7 tokens (    1,87 ms per token,   533,66 tokens per second)
0.01.116.943 I common_perf_print:        eval time =     502,86 ms /    63 runs   (    7,98 ms per token,   125,28 tokens per second)
0.01.116.943 I common_perf_print:       total time =     526,32 ms /    70 tokens
0.01.116.944 I common_perf_print: unaccounted time =       1,34 ms /   0,3 %      (total - sampling - prompt eval - eval) / (total)
0.01.116.944 I common_perf_print:    graphs reused =         62

real	0m1,181s
user	0m4,785s
sys	0m0,210s
  • q5_k:
+ ./bin/llama-completion -no-cnv --model ../models-mnt/qwen3/0.6B/ggml-model-q5_k.gguf -ngl 99 -c 1024 -s 1234 -n 64 --ignore-eos -p 'I believe the meaning of life is'
warning: no usable GPU found, --gpu-layers option will be ignored
warning: one possible reason is that llama.cpp was compiled without GPU support
warning: consult docs/build.md for compilation instructions
0.00.001.626 I llama_completion: llama backend init
0.00.001.630 I llama_completion: load the model and apply lora adapter, if any
0.00.001.637 I common_init_result: fitting params to device memory ...
0.00.001.638 I common_init_result: (for bugs during this step try to reproduce them with -fit off, or provide --verbose logs if the bug only occurs with -fit on)
0.00.191.223 I common_params_fit_impl: projected to use 839 MiB of host memory vs. 77898 MiB of total host memory
0.00.275.972 W load: control-looking token: 128247 '</s>' was not control-type; this is probably a bug in the model. its type will be overridden
0.00.347.684 W llama_context: n_ctx_seq (1024) < n_ctx_train (32768) -- the full capacity of the model will not be utilized
0.00.393.684 I common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
0.00.410.142 I llama_completion: llama threadpool init, n_threads = 8
0.00.410.164 I 
0.00.410.231 I system_info: n_threads = 8 (n_threads_batch = 8) / 16 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 | 
0.00.410.233 I 
0.00.410.308 I sampler seed: 1234
0.00.410.316 I sampler params: 
	repeat_last_n = 64, repeat_penalty = 1,000, frequency_penalty = 0,000, presence_penalty = 0,000
	dry_multiplier = 0,000, dry_base = 1,750, dry_allowed_length = 2, dry_penalty_last_n = -1
	top_k = 40, top_p = 0,950, min_p = 0,050, xtc_probability = 0,000, xtc_threshold = 0,100, typical_p = 1,000, top_n_sigma = -1,000, temp = 0,800
	mirostat = 0, mirostat_lr = 0,100, mirostat_ent = 5,000, adaptive_target = -1,000, adaptive_decay = 0,900
0.00.410.321 I sampler chain: logits -> logit-bias -> ?penalties -> ?dry -> ?top-n-sigma -> top-k -> ?typical -> top-p -> min-p -> ?xtc -> temp-ext -> dist 
0.00.410.322 I generate: n_ctx = 1024, n_batch = 2048, n_predict = 64, n_keep = 0
0.00.410.322 I 
I believe the meaning of life is to have fun, and to enjoy it. I think it is good to have fun, and to enjoy it, especially when it is the pursuit of something worthwhile. My life is filled with the pursuit of happiness. I spend a great deal of my time working on projects, reading books, and listening to music. I

0.01.001.021 I common_perf_print:    sampling time =       9,76 ms
0.01.001.022 I common_perf_print:    samplers time =       3,93 ms /    71 tokens
0.01.001.027 I common_perf_print:        load time =     180,11 ms
0.01.001.028 I common_perf_print: prompt eval time =      20,33 ms /     7 tokens (    2,90 ms per token,   344,27 tokens per second)
0.01.001.029 I common_perf_print:        eval time =     559,29 ms /    63 runs   (    8,88 ms per token,   112,64 tokens per second)
0.01.001.029 I common_perf_print:       total time =     590,89 ms /    70 tokens
0.01.001.030 I common_perf_print: unaccounted time =       1,51 ms /   0,3 %      (total - sampling - prompt eval - eval) / (total)
0.01.001.031 I common_perf_print:    graphs reused =         62

real	0m1,053s
user	0m5,198s
sys	0m0,128s
  • q6_k:
+ ./bin/llama-completion -no-cnv --model ../models-mnt/qwen3/0.6B/ggml-model-q6_k.gguf -ngl 99 -c 1024 -s 1234 -n 64 --ignore-eos -p 'I believe the meaning of life is'
warning: no usable GPU found, --gpu-layers option will be ignored
warning: one possible reason is that llama.cpp was compiled without GPU support
warning: consult docs/build.md for compilation instructions
0.00.002.554 I llama_completion: llama backend init
0.00.002.559 I llama_completion: load the model and apply lora adapter, if any
0.00.002.566 I common_init_result: fitting params to device memory ...
0.00.002.566 I common_init_result: (for bugs during this step try to reproduce them with -fit off, or provide --verbose logs if the bug only occurs with -fit on)
0.00.203.740 I common_params_fit_impl: projected to use 888 MiB of host memory vs. 77898 MiB of total host memory
0.00.300.657 W load: control-looking token: 128247 '</s>' was not control-type; this is probably a bug in the model. its type will be overridden
0.00.375.800 W llama_context: n_ctx_seq (1024) < n_ctx_train (32768) -- the full capacity of the model will not be utilized
0.00.420.335 I common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
0.00.437.505 I llama_completion: llama threadpool init, n_threads = 8
0.00.437.525 I 
0.00.437.593 I system_info: n_threads = 8 (n_threads_batch = 8) / 16 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 | 
0.00.437.596 I 
0.00.437.677 I sampler seed: 1234
0.00.437.687 I sampler params: 
	repeat_last_n = 64, repeat_penalty = 1,000, frequency_penalty = 0,000, presence_penalty = 0,000
	dry_multiplier = 0,000, dry_base = 1,750, dry_allowed_length = 2, dry_penalty_last_n = -1
	top_k = 40, top_p = 0,950, min_p = 0,050, xtc_probability = 0,000, xtc_threshold = 0,100, typical_p = 1,000, top_n_sigma = -1,000, temp = 0,800
	mirostat = 0, mirostat_lr = 0,100, mirostat_ent = 5,000, adaptive_target = -1,000, adaptive_decay = 0,900
0.00.437.692 I sampler chain: logits -> logit-bias -> ?penalties -> ?dry -> ?top-n-sigma -> top-k -> ?typical -> top-p -> min-p -> ?xtc -> temp-ext -> dist 
0.00.437.693 I generate: n_ctx = 1024, n_batch = 2048, n_predict = 64, n_keep = 0
0.00.437.693 I 
I believe the meaning of life is to give back, and I have given back to the world. My life is full of love, and I love love. I love the way the world is beautiful, and I love the people I love, and I love all who love me. I love the beauty of the world, and I love the love that

0.01.076.365 I common_perf_print:    sampling time =       9,47 ms
0.01.076.366 I common_perf_print:    samplers time =       3,84 ms /    71 tokens
0.01.076.370 I common_perf_print:        load time =     194,56 ms
0.01.076.372 I common_perf_print: prompt eval time =      16,94 ms /     7 tokens (    2,42 ms per token,   413,32 tokens per second)
0.01.076.374 I common_perf_print:        eval time =     610,85 ms /    63 runs   (    9,70 ms per token,   103,14 tokens per second)
0.01.076.374 I common_perf_print:       total time =     638,87 ms /    70 tokens
0.01.076.375 I common_perf_print: unaccounted time =       1,61 ms /   0,3 %      (total - sampling - prompt eval - eval) / (total)
0.01.076.375 I common_perf_print:    graphs reused =         62

real	0m1,133s
user	0m5,610s
sys	0m0,139s
  • save-load-state:

/**
* Tests whether the OpenAI TypeScript SDK handles SSE keepalive comments
* (`: keepalive\n\n`) injected without errors or text pollution.
*
* Use --words N to generate a large context that forces a long prefill phase,
* which is what actually triggers keepalive injection in practice.
*
* Run directly against llama.cpp (baseline):
* BASE_URL=http://localhost:15231/v1 bun index.ts --words 40000
*/
import OpenAI from "openai";
// ── args ──────────────────────────────────────────────────────────────────────
const args = process.argv.slice(2);
const BASE_URL = argValue("--base-url") ?? process.env.BASE_URL ?? "http://localhost:8888/v1";
const MODEL = argValue("--model") ?? process.env.MODEL ?? "local";
const wordCount = Number(argValue("--words") ?? "0");
// ── random word corpus ────────────────────────────────────────────────────────
// 200 common English words — repeated and shuffled to reach the requested count.
const CORPUS = [
"the","be","to","of","and","a","in","that","have","it","for","not","on","with",
"he","as","you","do","at","this","but","his","by","from","they","we","say","her",
"she","or","an","will","my","one","all","would","there","their","what","so","up",
"out","if","about","who","get","which","go","me","when","make","can","like","time",
"no","just","him","know","take","people","into","year","your","good","some","could",
"them","see","other","than","then","now","look","only","come","its","over","think",
"also","back","after","use","two","how","our","work","first","well","way","even",
"new","want","because","any","these","give","day","most","us","great","between",
"need","large","often","hand","high","place","hold","turn","help","small","home",
"read","set","move","play","spell","air","away","animal","house","point","page",
"letter","mother","answer","found","still","learn","plant","cover","food","sun",
"four","between","state","keep","eye","never","last","let","thought","city","tree",
"cross","farm","hard","start","might","story","saw","far","sea","draw","left",
"late","run","don","while","press","close","night","real","life","few","north",
"open","seem","together","next","white","children","begin","got","walk","example",
"ease","paper","group","always","music","those","both","mark","book","carry","took",
"science","eat","room","friend","began","idea","body","stop","once","book","hear",
];
function generateWords(n: number): string {
if (n === 0) return "";
const result: string[] = [];
while (result.length < n) {
// shuffle a copy and append
const shuffled = [...CORPUS].sort(() => Math.random() - 0.5);
result.push(...shuffled);
}
return result.slice(0, n).join(" ");
}
// ── build messages ────────────────────────────────────────────────────────────
const messages: OpenAI.ChatCompletionMessageParam[] = [];
if (wordCount > 0) {
const padding = generateWords(wordCount);
messages.push({
role: "system",
content:
`The following is context text (ignore its content, it is only here to ` +
`create a large prompt for testing purposes):\n\n${padding}`,
});
console.log(` padding : ${wordCount.toLocaleString()} words (~${Math.round(wordCount * 1.3).toLocaleString()} tokens estimated)`);
}
messages.push({
role: "user",
content: "Reply with exactly one sentence confirming you received this message.",
});
// ── client ────────────────────────────────────────────────────────────────────
console.log("SSE comment compatibility test");
console.log("─".repeat(50));
console.log(` base URL : ${BASE_URL}`);
console.log(` model : ${MODEL}`);
console.log(` messages : ${messages.length}`);
console.log();
const client = new OpenAI({
baseURL: BASE_URL,
apiKey: "test-key",
timeout: 30_000_000,
maxRetries: 0,
});
// ── fetch patch — observe raw SSE lines without interfering with the SDK ──────
let keepaliveComments = 0;
let otherComments = 0;
const originalFetch = globalThis.fetch;
// @ts-ignore — patching global fetch
globalThis.fetch = async (input: RequestInfo | URL, init?: RequestInit) => {
const response = await originalFetch(input, init);
const ct = response.headers.get("content-type") ?? "";
if (!ct.includes("text/event-stream") || !response.body) return response;
const [forObserver, forSdk] = response.body.tee();
(async () => {
const reader = forObserver.getReader();
const decoder = new TextDecoder();
let buffer = "";
try {
while (true) {
const { done, value } = await reader.read();
if (done) break;
buffer += decoder.decode(value, { stream: true });
const lines = buffer.split("\n");
buffer = lines.pop() ?? "";
for (const line of lines) {
if (line.startsWith(": keepalive")) {
keepaliveComments++;
console.log(`\n [raw SSE] keepalive comment #${keepaliveComments} (+${elapsed()}s)`);
} else if (line.startsWith(":")) {
otherComments++;
}
}
}
} catch { /* observer errors are non-fatal */ }
})();
return new Response(forSdk, {
status: response.status,
statusText: response.statusText,
headers: response.headers,
});
};
// ── stream ────────────────────────────────────────────────────────────────────
let chunkCount = 0;
let tokenCount = 0;
let fullText = "";
const start = Date.now();
try {
const stream = await client.chat.completions.create({
model: MODEL,
messages,
stream: true,
});
for await (const chunk of stream) {
chunkCount++;
const token = chunk.choices[0]?.delta?.content ?? "";
if (token) {
tokenCount++;
fullText += token;
process.stdout.write(token);
}
}
console.log("\n");
console.log("─".repeat(50));
console.log("Result: ✓ stream completed");
console.log(` elapsed : ${elapsed()}s`);
console.log(` SDK chunks : ${chunkCount}`);
console.log(` tokens received : ${tokenCount}`);
console.log(` keepalive comments : ${keepaliveComments}`);
console.log(` other comments : ${otherComments}`);
// Check 1: SDK handled comments without crashing
if (keepaliveComments > 0) {
console.log(" ✓ SDK handled keepalive comments without error");
} else {
console.log(" ~ no keepalive comments seen (direct connection, or interval not reached)");
}
// Check 2: comments did not bleed into assembled text
const polluted = fullText.includes(": keepalive") || fullText.includes(": keep");
if (polluted) {
console.log(" ✗ POLLUTED — keepalive content leaked into response text:");
console.log(` ${JSON.stringify(fullText.slice(0, 200))}`);
process.exit(1);
} else if (keepaliveComments > 0) {
console.log(" ✓ response text is clean — no keepalive content leaked");
}
} catch (err: unknown) {
console.log("\n");
console.log("─".repeat(50));
console.error(`Result: ✗ stream failed after ${elapsed()}s`);
if (err instanceof Error) {
let e: Error | undefined = err;
let depth = 0;
while (e) {
console.error(`${" ".repeat(depth)}${e.constructor.name}: ${e.message}`);
e = (e as any).cause instanceof Error ? (e as any).cause : undefined;
depth++;
}
}
console.log(` keepalive comments seen before failure: ${keepaliveComments}`);
process.exit(1);
}
// ── helpers ───────────────────────────────────────────────────────────────────
function elapsed(): string {
return ((Date.now() - start) / 1000).toFixed(1);
}
function argValue(flag: string): string | null {
const i = args.indexOf(flag);
return i !== -1 && args[i + 1] ? args[i + 1] : null;
}
{
"name": "sse-comment-test",
"version": "0.1.0",
"description": "Test whether the OpenAI TypeScript SDK handles SSE keepalive comments",
"type": "module",
"scripts": {
"start": "node --experimental-strip-types index.ts",
"bun": "bun index.ts"
},
"dependencies": {
"openai": "^4"
}
}
============================= test session starts ==============================
platform linux -- Python 3.12.3, pytest-8.3.5, pluggy-1.6.0 -- /root/llama.me/tools/server/tests/venv/bin/python3
cachedir: .pytest_cache
rootdir: /root/llama.me/tools/server/tests
configfile: pytest.ini
plugins: anyio-4.13.0
collecting ... collected 484 items / 199 deselected / 285 selected
unit/test_basic.py::test_server_start_simple PASSED [ 0%]
unit/test_basic.py::test_server_props PASSED [ 0%]
unit/test_basic.py::test_server_models PASSED [ 1%]
unit/test_basic.py::test_server_slots PASSED [ 1%]
unit/test_basic.py::test_load_split_model PASSED [ 1%]
unit/test_basic.py::test_no_webui PASSED [ 2%]
unit/test_basic.py::test_server_model_aliases_and_tags PASSED [ 2%]
unit/test_chat_completion.py::test_chat_completion[None-Book-Hey-8-But she couldn't-69-8-length-False-None] PASSED [ 2%]
unit/test_chat_completion.py::test_chat_completion[None-Book-Hey-8-But she couldn't-69-8-length-True-None] PASSED [ 3%]
unit/test_chat_completion.py::test_chat_completion[None-Book-What is the best book-8-(Suddenly)+|\\{ " Sarax.-77-8-length-False-None] PASSED [ 3%]
unit/test_chat_completion.py::test_chat_completion[None-Book-What is the best book-8-(Suddenly)+|\\{ " Sarax.-77-8-length-True-None] PASSED [ 3%]
unit/test_chat_completion.py::test_chat_completion[None-Book-What is the best book-8-(Suddenly)+|\\{ " Sarax.-77-8-length-True-chatml] PASSED [ 4%]
unit/test_chat_completion.py::test_chat_completion[None-Book-What is the best book-8-^ blue-23-8-length-True-This is not a chat template, it is] PASSED [ 4%]
unit/test_chat_completion.py::test_chat_completion[codellama70b-You are a coding assistant.-Write the fibonacci function in c++.-128-(Aside|she|felter|alonger)+-104-128-length-False-None] PASSED [ 4%]
unit/test_chat_completion.py::test_chat_completion[codellama70b-You are a coding assistant.-Write the fibonacci function in c++.-128-(Aside|she|felter|alonger)+-104-128-length-True-None] PASSED [ 5%]
unit/test_chat_completion.py::test_chat_completion[None-Book-user_prompt8-8-Whillicter-79-8-length-False-None] PASSED [ 5%]
unit/test_chat_completion.py::test_chat_completion[None-Book-user_prompt9-8-Whillicter-79-8-length-True-None] PASSED [ 5%]
unit/test_chat_completion.py::test_chat_completion_cached_tokens PASSED [ 6%]
unit/test_chat_completion.py::test_chat_completion_stream[Book-What is the best book-8-(Suddenly)+-77-8-length] PASSED [ 6%]
unit/test_chat_completion.py::test_chat_completion_stream[You are a coding assistant.-Write the fibonacci function in c++.-128-(Aside|she|felter|alonger)+-104-128-length] PASSED [ 7%]
unit/test_chat_completion.py::test_chat_completion_with_openai_library PASSED [ 7%]
unit/test_chat_completion.py::test_chat_template PASSED [ 7%]
unit/test_chat_completion.py::test_chat_template_assistant_prefill[Whill-Whill] PASSED [ 8%]
unit/test_chat_completion.py::test_chat_template_assistant_prefill[prefill1-Wh\n\nill] PASSED [ 8%]
unit/test_chat_completion.py::test_chat_template_continue_final_message_vllm_compat PASSED [ 8%]
unit/test_chat_completion.py::test_chat_template_continue_final_message_mutual_exclusion PASSED [ 9%]
unit/test_chat_completion.py::test_apply_chat_template PASSED [ 9%]
unit/test_chat_completion.py::test_completion_with_response_format[response_format0-6-"42"] PASSED [ 9%]
unit/test_chat_completion.py::test_completion_with_response_format[response_format1-10-[ -3000 ]] PASSED [ 10%]
unit/test_chat_completion.py::test_completion_with_response_format[response_format2-10-"foooooo"] PASSED [ 10%]
unit/test_chat_completion.py::test_completion_with_response_format[response_format3-10-(\\{|John)+] PASSED [ 10%]
unit/test_chat_completion.py::test_completion_with_response_format[response_format4-0-None] PASSED [ 11%]
unit/test_chat_completion.py::test_completion_with_response_format[response_format5-0-None] PASSED [ 11%]
unit/test_chat_completion.py::test_completion_with_response_format[response_format6-0-None] PASSED [ 11%]
unit/test_chat_completion.py::test_completion_with_response_format[response_format7-0-None] PASSED [ 12%]
unit/test_chat_completion.py::test_completion_with_json_schema[False-json_schema0-6-"42"] PASSED [ 12%]
unit/test_chat_completion.py::test_completion_with_json_schema[True-json_schema1-6-"42"] PASSED [ 12%]
unit/test_chat_completion.py::test_completion_with_grammar[False-root ::= "a"{5,5}-6-a{5,5}] PASSED [ 13%]
unit/test_chat_completion.py::test_completion_with_grammar[True-root ::= "a"{5,5}-6-a{5,5}] PASSED [ 13%]
unit/test_chat_completion.py::test_invalid_chat_completion_req[None] PASSED [ 14%]
unit/test_chat_completion.py::test_invalid_chat_completion_req[string] PASSED [ 14%]
unit/test_chat_completion.py::test_invalid_chat_completion_req[messages2] PASSED [ 14%]
unit/test_chat_completion.py::test_invalid_chat_completion_req[messages3] PASSED [ 15%]
unit/test_chat_completion.py::test_invalid_chat_completion_req[messages4] PASSED [ 15%]
unit/test_chat_completion.py::test_invalid_chat_completion_req[messages5] PASSED [ 15%]
unit/test_chat_completion.py::test_invalid_chat_completion_req[messages6] PASSED [ 16%]
unit/test_chat_completion.py::test_invalid_chat_completion_req[messages7] PASSED [ 16%]
unit/test_chat_completion.py::test_chat_completion_with_timings_per_token PASSED [ 16%]
unit/test_chat_completion.py::test_logprobs PASSED [ 17%]
unit/test_chat_completion.py::test_logprobs_stream PASSED [ 17%]
unit/test_chat_completion.py::test_logit_bias PASSED [ 17%]
unit/test_chat_completion.py::test_context_size_exceeded PASSED [ 18%]
unit/test_chat_completion.py::test_context_size_exceeded_stream PASSED [ 18%]
unit/test_chat_completion.py::test_return_progress[64-4-False] PASSED [ 18%]
unit/test_chat_completion.py::test_return_progress[64-2-True] PASSED [ 19%]
unit/test_chat_completion.py::test_chat_completions_multiple_choices PASSED [ 19%]
unit/test_compat_anthropic.py::test_anthropic_messages_basic PASSED [ 20%]
unit/test_compat_anthropic.py::test_anthropic_messages_with_system PASSED [ 20%]
unit/test_compat_anthropic.py::test_anthropic_messages_multipart_content PASSED [ 20%]
unit/test_compat_anthropic.py::test_anthropic_messages_conversation PASSED [ 21%]
unit/test_compat_anthropic.py::test_anthropic_messages_streaming PASSED [ 21%]
unit/test_compat_anthropic.py::test_anthropic_count_tokens PASSED [ 21%]
unit/test_compat_anthropic.py::test_anthropic_count_tokens_with_system PASSED [ 22%]
unit/test_compat_anthropic.py::test_anthropic_count_tokens_no_max_tokens PASSED [ 22%]
unit/test_compat_anthropic.py::test_anthropic_tool_use_basic PASSED [ 22%]
unit/test_compat_anthropic.py::test_anthropic_tool_result PASSED [ 23%]
unit/test_compat_anthropic.py::test_anthropic_tool_result_with_text PASSED [ 23%]
unit/test_compat_anthropic.py::test_anthropic_tool_result_error PASSED [ 23%]
unit/test_compat_anthropic.py::test_anthropic_tool_streaming PASSED [ 24%]
unit/test_compat_anthropic.py::test_anthropic_vision_format_accepted PASSED [ 24%]
unit/test_compat_anthropic.py::test_anthropic_vision_base64_with_multimodal_model PASSED [ 24%]
unit/test_compat_anthropic.py::test_anthropic_stop_sequences PASSED [ 25%]
unit/test_compat_anthropic.py::test_anthropic_temperature PASSED [ 25%]
unit/test_compat_anthropic.py::test_anthropic_top_p PASSED [ 25%]
unit/test_compat_anthropic.py::test_anthropic_top_k PASSED [ 26%]
unit/test_compat_anthropic.py::test_anthropic_missing_messages PASSED [ 26%]
unit/test_compat_anthropic.py::test_anthropic_empty_messages PASSED [ 27%]
unit/test_compat_anthropic.py::test_anthropic_streaming_content_block_indices PASSED [ 27%]
unit/test_compat_anthropic.py::test_anthropic_thinking PASSED [ 27%]
unit/test_compat_anthropic.py::test_anthropic_metadata PASSED [ 28%]
unit/test_compat_anthropic.py::test_anthropic_vs_openai_different_response_format PASSED [ 28%]
unit/test_compat_anthropic.py::test_anthropic_thinking_history_in_count_tokens PASSED [ 28%]
unit/test_compat_anthropic.py::test_anthropic_thinking_history_in_template PASSED [ 29%]
unit/test_compat_gcp.py::test_gcp_predict_camel_case PASSED [ 29%]
unit/test_compat_gcp.py::test_gcp_predict_multiple_instances PASSED [ 29%]
unit/test_compat_oai_responses.py::test_responses_with_openai_library PASSED [ 30%]
unit/test_compat_oai_responses.py::test_responses_stream_with_openai_library PASSED [ 30%]
unit/test_completion.py::test_completion[I believe the meaning of life is-8-(going|bed)+-18-8-False-False] PASSED [ 30%]
unit/test_completion.py::test_completion[Write a joke about AI from a very long prompt which will not be truncated-64-(princesses|everyone|kids|Anna|forest)+-46-64-False-True] PASSED [ 31%]
unit/test_completion.py::test_completion_stream[I believe the meaning of life is-8-(going|bed)+-18-8-False] PASSED [ 31%]
unit/test_completion.py::test_completion_stream[Write a joke about AI from a very long prompt which will not be truncated-64-(princesses|everyone|kids|Anna|forest)+-46-64-False] PASSED [ 31%]
unit/test_completion.py::test_completion_stream_vs_non_stream PASSED [ 32%]
unit/test_completion.py::test_completion_with_openai_library PASSED [ 32%]
unit/test_completion.py::test_completion_stream_with_openai_library PASSED [ 32%]
unit/test_completion.py::test_consistent_result_same_seed[1] PASSED [ 33%]
unit/test_completion.py::test_consistent_result_same_seed[2] PASSED [ 33%]
unit/test_completion.py::test_different_result_different_seed[1] PASSED [ 34%]
unit/test_completion.py::test_different_result_different_seed[2] PASSED [ 34%]
unit/test_completion.py::test_consistent_result_different_batch_size[0.0-16] PASSED [ 34%]
unit/test_completion.py::test_consistent_result_different_batch_size[0.0-32] PASSED [ 35%]
unit/test_completion.py::test_cache_vs_nocache_prompt SKIPPED (This ...) [ 35%]
unit/test_completion.py::test_nocache_long_input_prompt PASSED [ 35%]
unit/test_completion.py::test_json_prompt_no_mtmd PASSED [ 36%]
unit/test_completion.py::test_json_prompt_mtm_error_when_not_supported PASSED [ 36%]
unit/test_completion.py::test_completion_with_tokens_input PASSED [ 36%]
unit/test_completion.py::test_completion_parallel_slots[1-3] PASSED [ 37%]
unit/test_completion.py::test_completion_parallel_slots[2-2] PASSED [ 37%]
unit/test_completion.py::test_completion_parallel_slots[2-4] PASSED [ 37%]
unit/test_completion.py::test_completion_parallel_slots[4-2] PASSED [ 38%]
unit/test_completion.py::test_completion_parallel_slots[4-6] PASSED [ 38%]
unit/test_completion.py::test_completion_unified[256-4-n_predict_vals0-expected_success0] PASSED [ 38%]
unit/test_completion.py::test_completion_unified[256-4-n_predict_vals1-expected_success1] PASSED [ 39%]
unit/test_completion.py::test_completion_unified[256-4-n_predict_vals2-expected_success2] PASSED [ 39%]
unit/test_completion.py::test_completion_unified[256-4-n_predict_vals3-expected_success3] PASSED [ 40%]
unit/test_completion.py::test_completion_response_fields[I believe the meaning of life is-8-response_fields0] PASSED [ 40%]
unit/test_completion.py::test_completion_response_fields[I believe the meaning of life is-32-response_fields1] PASSED [ 40%]
unit/test_completion.py::test_n_probs PASSED [ 41%]
unit/test_completion.py::test_n_probs_stream PASSED [ 41%]
unit/test_completion.py::test_n_probs_post_sampling PASSED [ 41%]
unit/test_completion.py::test_n_probs_post_backend_sampling PASSED [ 42%]
unit/test_completion.py::test_logit_bias[False-False] PASSED [ 42%]
unit/test_completion.py::test_logit_bias[False-True] PASSED [ 42%]
unit/test_completion.py::test_logit_bias[True-False] PASSED [ 43%]
unit/test_completion.py::test_logit_bias[True-True] PASSED [ 43%]
unit/test_completion.py::test_cancel_request PASSED [ 43%]
unit/test_completion.py::test_completion_prompt_cache PASSED [ 44%]
unit/test_ctx_shift.py::test_ctx_shift_enabled PASSED [ 44%]
unit/test_ctx_shift.py::test_ctx_shift_disabled_short_prompt[64-64-False] PASSED [ 44%]
unit/test_ctx_shift.py::test_ctx_shift_disabled_short_prompt[-1-248-True] PASSED [ 45%]
unit/test_ctx_shift.py::test_ctx_shift_disabled_long_prompt PASSED [ 45%]
unit/test_ctx_shift.py::test_ctx_shift_disabled_stream PASSED [ 45%]
unit/test_embedding.py::test_embedding_single PASSED [ 46%]
unit/test_embedding.py::test_embedding_multiple PASSED [ 46%]
unit/test_embedding.py::test_embedding_multiple_with_fa PASSED [ 47%]
unit/test_embedding.py::test_embedding_mixed_input[-False] PASSED [ 47%]
unit/test_embedding.py::test_embedding_mixed_input[string-False] PASSED [ 47%]
unit/test_embedding.py::test_embedding_mixed_input[input2-False] PASSED [ 48%]
unit/test_embedding.py::test_embedding_mixed_input[input3-False] PASSED [ 48%]
unit/test_embedding.py::test_embedding_mixed_input[input4-True] PASSED [ 48%]
unit/test_embedding.py::test_embedding_mixed_input[input5-True] PASSED [ 49%]
unit/test_embedding.py::test_embedding_mixed_input[input6-True] PASSED [ 49%]
unit/test_embedding.py::test_embedding_mixed_input[input7-True] PASSED [ 49%]
unit/test_embedding.py::test_embedding_pooling_mean PASSED [ 50%]
unit/test_embedding.py::test_embedding_pooling_mean_multiple PASSED [ 50%]
unit/test_embedding.py::test_embedding_pooling_none PASSED [ 50%]
unit/test_embedding.py::test_embedding_pooling_none_oai PASSED [ 51%]
unit/test_embedding.py::test_embedding_openai_library_single PASSED [ 51%]
unit/test_embedding.py::test_embedding_openai_library_multiple PASSED [ 51%]
unit/test_embedding.py::test_embedding_error_prompt_too_long PASSED [ 52%]
unit/test_embedding.py::test_same_prompt_give_same_result PASSED [ 52%]
unit/test_embedding.py::test_embedding_usage_single[I believe the meaning of life is-9] PASSED [ 52%]
unit/test_embedding.py::test_embedding_usage_single[This is a test-6] PASSED [ 53%]
unit/test_embedding.py::test_embedding_usage_multiple PASSED [ 53%]
unit/test_embedding.py::test_embedding_openai_library_base64 PASSED [ 54%]
unit/test_ignore_eos.py::test_ignore_eos_populates_logit_bias PASSED [ 54%]
unit/test_ignore_eos.py::test_ignore_eos_false_no_logit_bias PASSED [ 54%]
unit/test_infill.py::test_infill_without_input_extra PASSED [ 55%]
unit/test_infill.py::test_infill_with_input_extra PASSED [ 55%]
unit/test_infill.py::test_invalid_input_extra_req[input_extra0] PASSED [ 55%]
unit/test_infill.py::test_invalid_input_extra_req[input_extra1] PASSED [ 56%]
unit/test_infill.py::test_invalid_input_extra_req[input_extra2] PASSED [ 56%]
unit/test_infill.py::test_invalid_input_extra_req[input_extra3] PASSED [ 56%]
unit/test_infill.py::test_invalid_input_extra_req[input_extra4] PASSED [ 57%]
unit/test_infill.py::test_with_qwen_model SKIPPED (skipping slow test) [ 57%]
unit/test_kv_keep_only_active.py::test_clear_and_restore PASSED [ 57%]
unit/test_kv_keep_only_active.py::test_disabled_with_flag PASSED [ 58%]
unit/test_lora.py::test_lora[0.0-(little|girl|three|years|old)+] PASSED [ 58%]
unit/test_lora.py::test_lora[1.0-(eye|love|glass|sun)+] PASSED [ 58%]
unit/test_lora.py::test_lora_per_request PASSED [ 59%]
unit/test_lora.py::test_with_big_model SKIPPED (skipping slow test) [ 59%]
unit/test_proxy.py::test_mcp_no_proxy PASSED [ 60%]
unit/test_proxy.py::test_mcp_proxy PASSED [ 60%]
unit/test_proxy.py::test_mcp_proxy_custom_port PASSED [ 60%]
unit/test_rerank.py::test_rerank PASSED [ 61%]
unit/test_rerank.py::test_rerank_tei_format PASSED [ 61%]
unit/test_rerank.py::test_invalid_rerank_req[documents0] PASSED [ 61%]
unit/test_rerank.py::test_invalid_rerank_req[None] PASSED [ 62%]
unit/test_rerank.py::test_invalid_rerank_req[123] PASSED [ 62%]
unit/test_rerank.py::test_invalid_rerank_req[documents3] PASSED [ 62%]
unit/test_rerank.py::test_rerank_usage[Machine learning is-A machine-Learning is-19] PASSED [ 63%]
unit/test_rerank.py::test_rerank_usage[Which city?-Machine learning is -Paris, capitale de la-26] PASSED [ 63%]
unit/test_rerank.py::test_rerank_top_n[None-4] PASSED [ 63%]
unit/test_rerank.py::test_rerank_top_n[2-2] PASSED [ 64%]
unit/test_rerank.py::test_rerank_top_n[4-4] PASSED [ 64%]
unit/test_rerank.py::test_rerank_top_n[99-4] PASSED [ 64%]
unit/test_rerank.py::test_rerank_tei_top_n[None-4] PASSED [ 65%]
unit/test_rerank.py::test_rerank_tei_top_n[2-2] PASSED [ 65%]
unit/test_rerank.py::test_rerank_tei_top_n[4-4] PASSED [ 65%]
unit/test_rerank.py::test_rerank_tei_top_n[99-4] PASSED [ 66%]
unit/test_router.py::test_router_props PASSED [ 66%]
unit/test_router.py::test_router_chat_completion_stream[ggml-org/tinygemma3-GGUF:Q8_0-True] PASSED [ 67%]
unit/test_router.py::test_router_chat_completion_stream[non-existent/model-False] PASSED [ 67%]
unit/test_router.py::test_router_unload_model PASSED [ 67%]
unit/test_router.py::test_router_models_max_evicts_lru PASSED [ 68%]
unit/test_router.py::test_router_no_models_autoload PASSED [ 68%]
unit/test_router.py::test_router_api_key_required PASSED [ 68%]
unit/test_router.py::test_router_reload_models PASSED [ 69%]
unit/test_security.py::test_access_public_endpoint[/health] PASSED [ 69%]
unit/test_security.py::test_access_public_endpoint[/models] PASSED [ 69%]
unit/test_security.py::test_access_static_assets_without_api_key PASSED [ 70%]
unit/test_security.py::test_incorrect_api_key[None] PASSED [ 70%]
unit/test_security.py::test_incorrect_api_key[invalid-key] PASSED [ 70%]
unit/test_security.py::test_correct_api_key PASSED [ 71%]
unit/test_security.py::test_correct_api_key_anthropic_header PASSED [ 71%]
unit/test_security.py::test_openai_library_correct_api_key PASSED [ 71%]
unit/test_security.py::test_cors_options[localhost-Access-Control-Allow-Origin-localhost] PASSED [ 72%]
unit/test_security.py::test_cors_options[web.mydomain.fr-Access-Control-Allow-Origin-web.mydomain.fr] PASSED [ 72%]
unit/test_security.py::test_cors_options[origin-Access-Control-Allow-Credentials-true] PASSED [ 72%]
unit/test_security.py::test_cors_options[web.mydomain.fr-Access-Control-Allow-Methods-GET, POST] PASSED [ 73%]
unit/test_security.py::test_cors_options[web.mydomain.fr-Access-Control-Allow-Headers-*] PASSED [ 73%]
unit/test_security.py::test_local_media_file[None-file://mtmd/test-1.jpeg-False] PASSED [ 74%]
unit/test_security.py::test_local_media_file[../../../tools-file://mtmd/test-1.jpeg-True] PASSED [ 74%]
unit/test_security.py::test_local_media_file[../../../tools-file:////mtmd//test-1.jpeg-True] PASSED [ 74%]
unit/test_security.py::test_local_media_file[../../../tools-file://mtmd/notfound.jpeg-False] PASSED [ 75%]
unit/test_security.py::test_local_media_file[../../../tools-file://../mtmd/test-1.jpeg-False] PASSED [ 75%]
unit/test_sleep.py::test_server_sleep PASSED [ 75%]
unit/test_slot_save.py::test_slot_save_restore PASSED [ 76%]
unit/test_slot_save.py::test_slot_erase PASSED [ 76%]
unit/test_speculative.py::test_with_and_without_draft PASSED [ 76%]
unit/test_speculative.py::test_different_draft_min_draft_max PASSED [ 77%]
unit/test_speculative.py::test_slot_ctx_not_exceeded PASSED [ 77%]
unit/test_speculative.py::test_with_ctx_shift PASSED [ 77%]
unit/test_speculative.py::test_multi_requests_parallel[1-2] PASSED [ 78%]
unit/test_speculative.py::test_multi_requests_parallel[2-2] PASSED [ 78%]
unit/test_template.py::test_reasoning[deepseek-ai-DeepSeek-R1-Distill-Qwen-32B-on-<think>\n-None] PASSED [ 78%]
unit/test_template.py::test_reasoning[deepseek-ai-DeepSeek-R1-Distill-Qwen-32B-on-<think>\n-tools1] PASSED [ 79%]
unit/test_template.py::test_reasoning[deepseek-ai-DeepSeek-R1-Distill-Qwen-32B-on-<think>\n-tools2] PASSED [ 79%]
unit/test_template.py::test_reasoning[deepseek-ai-DeepSeek-R1-Distill-Qwen-32B-auto-<think>\n-None] PASSED [ 80%]
unit/test_template.py::test_reasoning[deepseek-ai-DeepSeek-R1-Distill-Qwen-32B-auto-<think>\n-tools1] PASSED [ 80%]
unit/test_template.py::test_reasoning[deepseek-ai-DeepSeek-R1-Distill-Qwen-32B-auto-<think>\n-tools2] PASSED [ 80%]
unit/test_template.py::test_reasoning[deepseek-ai-DeepSeek-R1-Distill-Qwen-32B-off-<think>\n</think>-None] PASSED [ 81%]
unit/test_template.py::test_reasoning[deepseek-ai-DeepSeek-R1-Distill-Qwen-32B-off-<think>\n</think>-tools1] PASSED [ 81%]
unit/test_template.py::test_reasoning[deepseek-ai-DeepSeek-R1-Distill-Qwen-32B-off-<think>\n</think>-tools2] PASSED [ 81%]
unit/test_template.py::test_reasoning[Qwen-Qwen3-0.6B-auto-<|im_start|>assistant\n-None] PASSED [ 82%]
unit/test_template.py::test_reasoning[Qwen-Qwen3-0.6B-auto-<|im_start|>assistant\n-tools1] PASSED [ 82%]
unit/test_template.py::test_reasoning[Qwen-Qwen3-0.6B-auto-<|im_start|>assistant\n-tools2] PASSED [ 82%]
unit/test_template.py::test_reasoning[Qwen-Qwen3-0.6B-off-<|im_start|>assistant\n<think>\n\n</think>\n\n-None] PASSED [ 83%]
unit/test_template.py::test_reasoning[Qwen-Qwen3-0.6B-off-<|im_start|>assistant\n<think>\n\n</think>\n\n-tools1] PASSED [ 83%]
unit/test_template.py::test_reasoning[Qwen-Qwen3-0.6B-off-<|im_start|>assistant\n<think>\n\n</think>\n\n-tools2] PASSED [ 83%]
unit/test_template.py::test_reasoning[Qwen-QwQ-32B-auto-<|im_start|>assistant\n<think>\n-None] PASSED [ 84%]
unit/test_template.py::test_reasoning[Qwen-QwQ-32B-auto-<|im_start|>assistant\n<think>\n-tools1] PASSED [ 84%]
unit/test_template.py::test_reasoning[Qwen-QwQ-32B-auto-<|im_start|>assistant\n<think>\n-tools2] PASSED [ 84%]
unit/test_template.py::test_reasoning[Qwen-QwQ-32B-off-<|im_start|>assistant\n<think>\n</think>-None] PASSED [ 85%]
unit/test_template.py::test_reasoning[Qwen-QwQ-32B-off-<|im_start|>assistant\n<think>\n</think>-tools1] PASSED [ 85%]
unit/test_template.py::test_reasoning[Qwen-QwQ-32B-off-<|im_start|>assistant\n<think>\n</think>-tools2] PASSED [ 85%]
unit/test_template.py::test_reasoning[CohereForAI-c4ai-command-r7b-12-2024-tool_use-auto-<|START_OF_TURN_TOKEN|><|CHATBOT_TOKEN|>-None] PASSED [ 86%]
unit/test_template.py::test_reasoning[CohereForAI-c4ai-command-r7b-12-2024-tool_use-auto-<|START_OF_TURN_TOKEN|><|CHATBOT_TOKEN|>-tools1] PASSED [ 86%]
unit/test_template.py::test_reasoning[CohereForAI-c4ai-command-r7b-12-2024-tool_use-auto-<|START_OF_TURN_TOKEN|><|CHATBOT_TOKEN|>-tools2] PASSED [ 87%]
unit/test_template.py::test_reasoning[CohereForAI-c4ai-command-r7b-12-2024-tool_use-off-<|START_OF_TURN_TOKEN|><|CHATBOT_TOKEN|><|START_THINKING|><|END_THINKING|>-None] PASSED [ 87%]
unit/test_template.py::test_reasoning[CohereForAI-c4ai-command-r7b-12-2024-tool_use-off-<|START_OF_TURN_TOKEN|><|CHATBOT_TOKEN|><|START_THINKING|><|END_THINKING|>-tools1] PASSED [ 87%]
unit/test_template.py::test_reasoning[CohereForAI-c4ai-command-r7b-12-2024-tool_use-off-<|START_OF_TURN_TOKEN|><|CHATBOT_TOKEN|><|START_THINKING|><|END_THINKING|>-tools2] PASSED [ 88%]
unit/test_template.py::test_date_inside_prompt[meta-llama-Llama-3.3-70B-Instruct-%d %b %Y-None] PASSED [ 88%]
unit/test_template.py::test_date_inside_prompt[meta-llama-Llama-3.3-70B-Instruct-%d %b %Y-tools1] PASSED [ 88%]
unit/test_template.py::test_date_inside_prompt[meta-llama-Llama-3.3-70B-Instruct-%d %b %Y-tools2] PASSED [ 89%]
unit/test_template.py::test_date_inside_prompt[fireworks-ai-llama-3-firefunction-v2-%b %d %Y-None] PASSED [ 89%]
unit/test_template.py::test_date_inside_prompt[fireworks-ai-llama-3-firefunction-v2-%b %d %Y-tools1] PASSED [ 89%]
unit/test_template.py::test_date_inside_prompt[fireworks-ai-llama-3-firefunction-v2-%b %d %Y-tools2] PASSED [ 90%]
unit/test_template.py::test_add_generation_prompt[meta-llama-Llama-3.3-70B-Instruct-<|start_header_id|>assistant<|end_header_id|>-False] PASSED [ 90%]
unit/test_template.py::test_add_generation_prompt[meta-llama-Llama-3.3-70B-Instruct-<|start_header_id|>assistant<|end_header_id|>-True] PASSED [ 90%]
unit/test_tokenize.py::test_tokenize_detokenize PASSED [ 91%]
unit/test_tokenize.py::test_tokenize_with_bos PASSED [ 91%]
unit/test_tokenize.py::test_tokenize_with_pieces PASSED [ 91%]
unit/test_tool_call.py::test_completion_without_tool_call_fast[meta-llama-Llama-3.3-70B-Instruct-128-tools0-None-CompletionMode.NORMAL] PASSED [ 92%]
unit/test_tool_call.py::test_completion_without_tool_call_fast[meta-llama-Llama-3.3-70B-Instruct-128-tools0-None-CompletionMode.STREAMED] PASSED [ 92%]
unit/test_tool_call.py::test_completion_without_tool_call_fast[meta-llama-Llama-3.3-70B-Instruct-128-tools1-None-CompletionMode.NORMAL] PASSED [ 92%]
unit/test_tool_call.py::test_completion_without_tool_call_fast[meta-llama-Llama-3.3-70B-Instruct-128-tools1-None-CompletionMode.STREAMED] PASSED [ 93%]
unit/test_tool_call.py::test_completion_without_tool_call_fast[meta-llama-Llama-3.3-70B-Instruct-128-tools2-none-CompletionMode.NORMAL] PASSED [ 93%]
unit/test_tool_call.py::test_completion_without_tool_call_fast[meta-llama-Llama-3.3-70B-Instruct-128-tools2-none-CompletionMode.STREAMED] PASSED [ 94%]
unit/test_vision_api.py::test_models_supports_multimodal_capability PASSED [ 94%]
unit/test_vision_api.py::test_v1_models_supports_multimodal_capability PASSED [ 94%]
unit/test_vision_api.py::test_vision_chat_completion[What is this:\n-IMG_URL_0-True-(cat)+] PASSED [ 95%]
unit/test_vision_api.py::test_vision_chat_completion[What is this:\n-IMG_BASE64_URI_0-True-(cat)+] PASSED [ 95%]
unit/test_vision_api.py::test_vision_chat_completion[What is this:\n-IMG_URL_1-True-(frog)+] PASSED [ 95%]
unit/test_vision_api.py::test_vision_chat_completion[Test test\n-IMG_URL_1-True-(frog)+] PASSED [ 96%]
unit/test_vision_api.py::test_vision_chat_completion[What is this:\n-malformed-False-None] PASSED [ 96%]
unit/test_vision_api.py::test_vision_chat_completion[What is this:\n-https://google.com/404-False-None] PASSED [ 96%]
unit/test_vision_api.py::test_vision_chat_completion[What is this:\n-https://ggml.ai-False-None] PASSED [ 97%]
unit/test_vision_api.py::test_vision_completion[What is this: <__media__>\n-IMG_BASE64_0-True-(cat)+] PASSED [ 97%]
unit/test_vision_api.py::test_vision_completion[What is this: <__media__>\n-IMG_BASE64_1-True-(frog)+] PASSED [ 97%]
unit/test_vision_api.py::test_vision_completion[What is this: <__media__>\n-malformed-False-None] PASSED [ 98%]
unit/test_vision_api.py::test_vision_completion[What is this:\n--False-None] PASSED [ 98%]
unit/test_vision_api.py::test_vision_embeddings[What is this: <__media__>\n-IMG_BASE64_0-True] PASSED [ 98%]
unit/test_vision_api.py::test_vision_embeddings[What is this: <__media__>\n-IMG_BASE64_1-True] PASSED [ 99%]
unit/test_vision_api.py::test_vision_embeddings[What is this: <__media__>\n-malformed-False] PASSED [ 99%]
unit/test_vision_api.py::test_vision_embeddings[What is this:\n-base64-False] PASSED [100%]
========== 282 passed, 3 skipped, 199 deselected in 455.53s (0:07:35) ==========
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment