Kokoro is an open-weight TTS model with 82 million parameters. Despite its lightweight architecture, it delivers comparable quality to larger models while being significantly faster and more cost-efficient. With Apache-licensed weights, Kokoro can be deployed anywhere from production environments to personal projects.
ubergarm
efemaer
/ kokoro-v1.0-benchmark.md
Last active
February 17, 2025 15:33
Kokoro v1 Benchmark (PyTorch/ONNX, CPU/GPU)
Relax, I only have one Sunday to work on idea, literally my weekend project. So I tried Deepseek to see if it can help. Surprisingly, it works and it saves me another weekend...
Just chat.deepseek.com (cost = free) with prompts adapted from this gist.
tristandruyen
/ calibration_data_v5_rc.txt
Last active
June 10, 2025 02:08
— forked from bartowski1182/calibration_datav3.txt
Adapted from bartowskis v3, added more languages for sparse moe models like qwen 57B-A14B. Calibration data provided by Dampf, combines his own efforts on top of Kalomaze's. Used for calibrating GGUF imatrix files
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| =========== | |
| ; A072257: a(n) = ((6*n-17)*4^n - 1)/3. | |
| ; -6,-15,-27,21,597,4437,25941,136533,677205,3233109,15029589,68506965,307582293,1364546901,5995058517,26127717717,113100805461,486762960213,2084490794325,8887718991189,37749899220309,159795689903445,674367131702613,2838206015165781,11915774014084437,49914895870022997,208666782734832981,870695927958295893,3626898899909039445,15084056351939581269,62642068416972019029,259791645704742851925,1076060070966390510933,4451814236455238456661,18397552756179659478357,75951394266153460520277,313250310030353132508501,1290780171984369691743573,5314236415389307413812565,21861408571364544242603349,89863485924687435319825749,369125350255666774676952405,1515187027250335232298407253,6215490613912013463556019541,25480932475290743991673640277,104399609979733736516492809557,427501960233217988265164232021,1749621922190004121857428903253,7156944013788545162616803513685,29261601355268295351215565657429,119581706621529640207855669040469,488468031287944396043396301804885,1994436944359 |
bartowski1182
/ calibration_datav3.txt
Last active
June 10, 2025 02:08
Calibration data provided by Dampf, combines his own efforts on top of Kalomaze's. Used for calibrating GGUF imatrix files
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| In addition to a significant decrease in hepatic lipid accumulation in the IOE group, which inhibited energy intake by propionate enrichment, hepatic lipids were also significantly reduced in the mice in the IOP group, which was largely enriched with butyrate. Compared with the IOE group, IOP had a stronger regulatory effect on hepatic metabolism and triglyceride metabolism and higher levels of TCA cycle in the host. In addition, butyrate has the ability to promote browning of white adipose tissue (WAT) to brown adipose tissue (BAT).^[@ref39],[@ref40]^ WAT stores energy, whereas BAT uses energy for heating and consequently host energy expenditure increases.^[@ref41],[@ref42]^ However, adipose tissue weight does not change after WAT browning.^[@ref43]^ Therefore, the weight of adipose tissue of mice in the IOP group dominated by butyrate was greater than that of the mice in the IOE group dominated by propionate. | |
| In conclusion ([Figure [7](#fig7){ref-type="fig"}](#fig7){ref-type="fig"}C), the improvement of ob |
Good question! I am collecting human data on how quantization affects outputs. See here for more information: ggml-org/llama.cpp#5962
In the meantime, use the largest that fully fits in your GPU. If you can comfortably fit Q4_K_S, try using a model with more parameters.
See the wiki upstream: https://github.com/ggerganov/llama.cpp/wiki/Feature-matrix
HugsLibRecordKeeper
/ output_log.txt
Created
December 10, 2020 00:17
Rimworld output log published using HugsLib
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Log uploaded on Thursday, December 10, 2020, 12:16:49 AM | |
| Loaded mods: | |
| Harmony(brrainz.harmony)[mv:1.0.4.0]: 0Harmony(2.0.2), HarmonyMod(1.0.4) | |
| Core(Ludeon.RimWorld): (no assemblies) | |
| SRTS Expanded(smashphil.neceros.srtsexpanded)[mv:1.4.6]: 0Harmony(av:2.0.2,fv:1.2.0.1), SRTS(1.0.0) | |
| Royalty(Ludeon.RimWorld.Royalty): (no assemblies) | |
| HugsLib(UnlimitedHugs.HugsLib)[ov:8.0.1]: 0Harmony(av:2.0.2,fv:1.2.0.1), HugsLib(av:1.0.0,fv:8.0.1) | |
| JecsTools (Unofficial)(jecrell.jecstools)[mv:1.1.2.2]: 0JecsTools(1.1.2.2), AbilityUser(1.1.2.2), AbilityUserAI(1.1.2.2), CompActivatableEffect(1.1.2.2), CompAnimated(1.1.2.2), CompBalloon(1.1.2.2), CompBigBox(1.1.2.2), CompDeflector(1.1.2.2), CompDelayedSpawner(1.1.2.2), CompExtraSounds(1.1.2.2), CompInstalledPart(1.1.2.2), CompLumbering(1.1.2.2), CompOverlays(1.1.2.2), CompOversizedWeapon(1.1.2.2), CompSlotLoadable(1.1.2.2), CompToggleDef(1.1.2.2), CompVehicle(1.1.2.1), PawnShields(1.1.2.2), ThinkNodes(1.1.2.2) | |
| FSharp.Core(latta.fsharp.core)[mv:4.8.2]: FSharp.Core(av:4.7.0,fv:4.700.2 |
HugsLibRecordKeeper
/ output_log.txt
Created
November 16, 2020 07:10
Rimworld output log published using HugsLib
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Log uploaded on Monday, November 16, 2020, 2:10:08 AM | |
| Loaded mods: | |
| Harmony(brrainz.harmony)[mv:1.0.4.0]: 0Harmony(2.0.2), HarmonyMod(1.0.4) | |
| Core(Ludeon.RimWorld): (no assemblies) | |
| Royalty(Ludeon.RimWorld.Royalty): (no assemblies) | |
| HugsLib(UnlimitedHugs.HugsLib)[ov:8.0.1]: 0Harmony(av:2.0.2,fv:1.2.0.1), HugsLib(av:1.0.0,fv:8.0.1) | |
| KanbanStockpile(ubergarm.kanbanstockpile): 0Harmony(av:2.0.2,fv:2.0.4), 0MultiplayerAPI(av:0.2.0,fv:0.1.0), KanbanStockpile(1.0.7623.32571) | |
| Active Harmony patches: | |
| DebugWindowsOpener.DevToolStarterOnGUI: TRANS: HugsLib.Patches.DevToolStarterOnGUI_Patch.ExtendButtonsWindow |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| """ | |
| WARNING: dont use loguru, use structlog | |
| https://gist.github.com/nkhitrov/38adbb314f0d35371eba4ffb8f27078f | |
| Configure handlers and formats for application loggers. | |
| """ | |
| import logging | |
| import sys | |
| from pprint import pformat |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| # If you don‘t want to build it youself, you can try `docker pull killercai/postgres`. | |
| FROM healthcheck/postgres:latest | |
| # China debian mirror | |
| RUN sed -i s@/deb.debian.org/@/mirrors.aliyun.com/@g /etc/apt/sources.list | |
| RUN apt-get clean && apt-get update | |
| RUN apt-get install -y wget git build-essential libpq-dev python-dev postgresql-server-dev-all | |
| # SCWS (Simple Chinese Word Segmentation library) | |
| RUN cd /tmp && wget -q -O - http://www.xunsearch.com/scws/down/scws-1.2.1.tar.bz2 | tar xjf - && cd scws-1.2.1 && ./configure && make install | |
| # zhpaser (postgres plugin) |
ZijiaLewisLu
/ Tricks to Speed Up Data Loading with PyTorch.md
Last active
June 1, 2025 00:35
Tricks to Speed Up Data Loading with PyTorch
In most of deep learning projects, the training scripts always start with lines to load in data, which can easily take a handful minutes. Only after data ready can start testing my buggy code. It is so frustratingly often that I wait for ten minutes just to find I made a stupid typo, then I have to restart and wait for another ten minutes hoping no other typos are made.
In order to make my life easy, I devote lots of effort to reduce the overhead of I/O loading. Here I list some useful tricks I found and hope they also save you some time.
-
use Numpy Memmap to load array and say goodbye to HDF5.
I used to relay on HDF5 to read/write data, especially when loading only sub-part of all data. Yet that was before I realized how fast and charming Numpy Memmapfile is. In short, Memmapfile does not load in the whole array at open, and only later "lazily" load in the parts that are required for real operations.
Sometimes I may want to copy the full array to memory at once, as it makes later operations
NewerOlder