Goal: run unsloth/Qwen3.6-35B-A3B-MTP-GGUF (Q4_K_XL) entirely on the GPU with the
largest possible context, and find the fastest generation settings — including
the model's built-in MTP (Multi-Token Prediction) self-speculation.
This is the MoE sibling of the dense 27B (see Benchmarks.md): 35.5 B total
params but only ~3 B active per token, so it generates ~2.4× faster while
still fitting the full 256k context on 24 GB.
Dataframe Oriented Programming: https://csvbase.com/blog/1
Accessing up to date data quickly and easily (even on phone) and pulling decision making information out of it constantly should be effortless.
Something like the stories of the old APL mainframe environment.
| NB. What the hell is a fork, a hook and cap? | |
| NB. a is a simple array | |
| a =: 1 2 3 4 5 6 7 8 9 10 | |
| NB. What is the average of a? | |
| NB. sum divided-by count | |
| (+/ a) % (# a) | |
| 5.5 |
| NB. parallel each: peach | |
| NB. https://code.jsoftware.com/wiki/NYCJUG/2022-11-08#Work_with_Multi-threading | |
| echo 'Must be at least running J904' | |
| echo JLIB | |
| nproc =: {{ {. 8 T. '' }} | |
| createThreads=: 3 : '{{ 0 T. '''' }} ^:({. 8 T. '''')]''''' |
| #!/usr/bin/env python3 | |
| from urllib.parse import quote | |
| import subprocess | |
| def wttr(location): | |
| c = "curl wttr.in/" + quote(location) + '?0TQ' | |
| r = subprocess.run(c, shell=True, stdout=subprocess.PIPE) | |
| l = r.stdout.decode('utf-8').split('\n') |
| NB. `select a,b,c,d where a=5 and b>7 order by c desc limit 10` | |
| NB. TODO: read data from parquet with https://github.com/AshyIsMe/JArrow | |
| NB. Something like: 'a b c d'=: 'a';'b';'c';'d' readParquet 'foo.parquet' | |
| a=:?1e6$10 NB. 1e6 random ints 0to9 inclusive | |
| b=:?1e6$10 |
| 0 : 0 | |
| Can you crack Aaron’s password hash? He seems to like | |
| simple passwords. I’m sure he’ll use his name and birthday | in it. Hint: Aaron writes important dates as YYYYMMDD | |
| rather than YYYY-MM-DD or any other special character | separator. Hash: 7f4986da7d7b52fa81f98278e6ec9dcb. | |
| Author: moat, Pacific Northwest National Laboratory | |
| ) | |
| y=:":"0 (1900 + i.120) | |
| m=:(i.11) { ('0' ,"1 ":"0 (1+i.9)) , ('1' ,"1 ":"0 (1+i.9)) | |
| d=:('0' ,"1 ":"0 (1+i.9)) , (":"0) 10+i.22 |
| NB. jgl2 animation demo for j | |
| NB. Based on https://code.jsoftware.com/wiki/Guides/Window_Driver/Animation/Animated_JGL2 | |
| NB. by (c) 2019 by michal j wallace http://tangentstorm.com/ | |
| NB. free for use under the MIT license. https://opensource.org/licenses/MIT | |
| load 'gl2' | |
| coinsert 'jgl2' | |
| close_handler =: 3 : 0 |
| { pkgs ? import <nixpkgs> {} }: | |
| pkgs.mkShell { | |
| buildInputs = [ | |
| pkgs.gcc9 | |
| pkgs.gmp | |
| pkgs.primesieve | |
| ]; | |
| } | |
| NB. http://oeis.org/A342260 | |
| a =: verb : '0 } I. y = +/"1 (y-1) = y #.inv ^&2 (i.y^(y+1))' "0 | |
| a 2 | |
| 3 | |
| a 3 | |
| 31 | |
| a 4 | |
| 217 | |
| a 5 |