Commits · 7140c014058f9fa2a75ff315af6af9a59efcfe5f · OpenDAS / bitsandbytes

12 Apr, 2023 2 commits
- Refactored simulated fp8 modules into research.nn. · dd562c24
  Tim Dettmers authored Apr 12, 2023
  
  dd562c24
- Refactored triton into its own folder. Refactored fp8 matmuls. · ec1ea637
  Tim Dettmers authored Apr 12, 2023
  
  ec1ea637
11 Apr, 2023 3 commits
- Fixed CUDA Conda PyTorch 2.0 issues. · 4cd63def
  Tim Dettmers authored Apr 11, 2023
  
  4cd63def
- Fixed bug where beta2 was not passed into Lion 32-bit. · 2eb31083
  Tim Dettmers authored Apr 11, 2023
  
  2eb31083
- Fixed noisy tests for 8-bit Lion. · 792af5c8
  Tim Dettmers authored Apr 11, 2023
  
  792af5c8
01 Apr, 2023 1 commit
- cleaning and refactor · 7f87ba83
  Mitchell Wortsman authored Apr 01, 2023
  
  7f87ba83
31 Mar, 2023 2 commits
- Added triton test. · 30d21d58
  Tim Dettmers authored Mar 31, 2023
  
  30d21d58
- Added first triton test. · a13a522c
  Tim Dettmers authored Mar 31, 2023
  
  a13a522c
29 Mar, 2023 2 commits
- test · b373034e
  Mitchell Wortsman authored Mar 29, 2023
  
  b373034e
- triton-v1 · 5f3d9ada
  Mitchell Wortsman authored Mar 29, 2023
  
  5f3d9ada
22 Mar, 2023 1 commit
- add some code in test_optim.py, although it seems to be failing · a43cd200
  Phil Wang authored Mar 22, 2023
  
  a43cd200
21 Mar, 2023 1 commit
- Add force_no_igemmlt to test params · dcecbb26
  Max Ryabinin authored Mar 22, 2023
  
  dcecbb26
09 Mar, 2023 2 commits
- forget about tests for now, will test live on local enwik8 training · 8de29fc3
  Phil Wang authored Mar 09, 2023
  
  8de29fc3
- do a bunch of typical bookkeeping before getting to main lion logic · cb4c3c8c
  Phil Wang authored Mar 09, 2023
  
  cb4c3c8c
25 Feb, 2023 1 commit
- Handle more cases in test_linear_serialization · ac3ab281
  Max Ryabinin authored Feb 25, 2023
  
  ac3ab281
23 Feb, 2023 1 commit
- Added matmul_mixed. · c5c38ca1
  Tim Dettmers authored Feb 23, 2023
  
  c5c38ca1
21 Feb, 2023 1 commit
- [WIP] Implement proper serialization of Linear8bitLt · 58b09ee1
  Max Ryabinin authored Feb 21, 2023
  
  58b09ee1
14 Feb, 2023 4 commits
- Added more blocksizes for stochastic rounding; fixed dequant blocksize. · 2489d819
  Tim Dettmers authored Feb 14, 2023
  
  2489d819
- Fixed LinearFP8 and added tests. · 2dfa3ce1
  Tim Dettmers authored Feb 13, 2023
  
  2dfa3ce1
- Added forward/backward tests; removed bias. · ca323658
  Tim Dettmers authored Feb 13, 2023
  
  ca323658
- Added fp8 simulation layer. · 6bdb6c35
  Tim Dettmers authored Feb 13, 2023
  
  6bdb6c35
02 Feb, 2023 1 commit
- Added Int8 matmul support for all GPUs. Full backward support. · de535889
  Tim Dettmers authored Feb 01, 2023
  
  de535889
29 Jan, 2023 1 commit
- Added outlier detector and fake quantization layer. · c9f50506
  Tim Dettmers authored Jan 28, 2023
  
  c9f50506
02 Jan, 2023 1 commit
- CUDASetup only executed once + fixed circular import. · 336e2469
  Tim Dettmers authored Jan 02, 2023
  
  336e2469
19 Nov, 2022 1 commit
- Fixed k-bit quantization maps. · eb028e6e
  Tim Dettmers authored Nov 19, 2022
  
  eb028e6e
08 Nov, 2022 1 commit
- Fixed bug in cpu quant; faster GPU dequant. · 08fa2e7b
  Tim Dettmers authored Nov 07, 2022
  
  08fa2e7b
07 Nov, 2022 2 commits
- Fixed blockwise test and logic. · e0e697b1
  Tim Dettmers authored Nov 06, 2022
  
  e0e697b1
- Added blocksizes 2048, 1024, and 512 to blockwise quant. · 6bc2b992
  Tim Dettmers authored Nov 06, 2022
  
  6bc2b992
06 Nov, 2022 3 commits
- Added k<256 quantile estimate. · 2f2063ba
  Tim Dettmers authored Nov 06, 2022
  
  2f2063ba
- Added k-bit fp8 map. · 98cbc4bc
  Tim Dettmers authored Nov 06, 2022
  
  98cbc4bc
- Added k-bit linear quantization. · caf18325
  Tim Dettmers authored Nov 06, 2022
  
  caf18325
04 Nov, 2022 1 commit
- Added FP8 quantization map. · 1efb87d8
  Tim Dettmers authored Nov 03, 2022
  
  1efb87d8
27 Oct, 2022 3 commits
- Sort imports · 7a3c9af0
  Tom Aarsen authored Oct 27, 2022
```
Via isort
```
  7a3c9af0
- Simplify statements into equivalent, modern variants · 0b078403
  Tom Aarsen authored Oct 27, 2022
```
via pyupgrade --py37-plus. The changes e.g. are subclassing from object, calling super() with super(ThisClass, self), or old-style syntax formatting.
```
  0b078403
- Remove trailing whitespace & ensure newline at EOF · 1eec77d3
  Tom Aarsen authored Oct 27, 2022
  
  1eec77d3
25 Oct, 2022 1 commit
- Added CUDA SETUP instruction generator. · a371be30
  Tim Dettmers authored Oct 25, 2022
  
  a371be30
24 Oct, 2022 1 commit
- Isolated CUDASetup logging; all tests green. · df86625a
  Tim Dettmers authored Oct 24, 2022
  
  df86625a
20 Sep, 2022 3 commits
- try fp32 · 76ce9aa6
  justheuristic authored Sep 20, 2022
  
  76ce9aa6
- set threshold · 292a4787
  Tim Dettmers authored Sep 20, 2022
  
  292a4787
- review · a07825ac
  justheuristic authored Sep 20, 2022
  
  a07825ac