Commits

Breadcrumbs

History for

onmain

User selector

All users

Datepicker

All time

Commit History

Commits on Mar 9, 2024

• Native support for Bfloat16 for 1x16 and 2x8 CUDA kernels (Vahe1994#41)

  Show description for 2f568b3
  BlackSamorez

  authored
  2f568b3

Commits on Mar 6, 2024

• no dev

  BlackSamorez

  committed
  28c1433

• numba kernel no print

  BlackSamorez

  committed
  57df022

• kernel selection warning

  BlackSamorez

  committed
  1413bca

Commits on Mar 5, 2024

• prod

  BlackSamorez

  committed
  66bf08c

• better CPUU kernel case

  BlackSamorez

  committed
  5d6aed0

• Dynamic matmul/matvec

  BlackSamorez

  committed
  debea84

• two ops

  BlackSamorez

  committed
  596580e

• Dev version

  BlackSamorez

  committed
  a8e936a

Commits on Feb 22, 2024

• version requirements bump

  BlackSamorez

  committed
  08a5fa4

• 1.0.3

  BlackSamorez

  committed
  9567518

• Merge branch 'main' into compile

  BlackSamorez

  committed
  5baa68b

• Marked new inference version as dev

  BlackSamorez

  committed
  36dd39d

• Library impls for cuda kernels

  BlackSamorez

  committed
  a93c551

Commits on Feb 21, 2024

• cuda streams

  BlackSamorez

  committed
  4810b6c

Commits on Feb 20, 2024

• Merge branch 'backward' of github.com:Vahe1994/AQLM into backward

  BlackSamorez

  committed
  abbb0d9

• 1.0.2

  BlackSamorez

  committed
  1b01426

• Update inference_lib/src/aqlm/inference_kernels/triton_kernel.py

  Show description for 5752ee8
  

  BlackSamorez

  and

  justheuristic

  authored
  5752ee8

• python 3.8 support

  BlackSamorez

  committed
  898ff08

• isort

  BlackSamorez

  committed
  ca95dc4

Commits on Feb 19, 2024

• Merge branch 'main' into backward

  BlackSamorez

  committed
  acfee0e

Commits on Feb 18, 2024

• dequantization kernel

  BlackSamorez

  committed
  2fad09d

• real default

  BlackSamorez

  committed
  c2847cf

• clone

  BlackSamorez

  committed
  a7d5627

• correct backward pass selector

  BlackSamorez

  committed
  0412546

• device guards

  BlackSamorez

  committed
  8d81cdb

• comment

  BlackSamorez

  committed
  4b08a7e

• Optimize for training wrapper

  BlackSamorez

  committed
  672a665

• one-tiem kernel selection

  BlackSamorez

  committed
  6ba7711

• moved bias logic to cpp

  BlackSamorez

  committed
  16e9858

Commits on Feb 17, 2024

• cheaper init

  BlackSamorez

  committed
  f57778b

• a

  BlackSamorez

  committed
  16d248b

• prettier code

  BlackSamorez

  committed
  9465794

• triton kernel fix (still not working)

  BlackSamorez

  committed
  43f1c3a

• backward pass

  BlackSamorez

  committed
  ab842da

Pagination

PreviousNext