Commits · 6c6e5fcb5355c2d0fb499a75124b959b1bbc3d0c · OpenDAS / bitsandbytes

14 Jul, 2023 11 commits
- Added changelog entry. · 6c6e5fcb
  Tim Dettmers authored Jul 14, 2023
  
  6c6e5fcb
- Polished CUDA SETUP replacement and added docs. · 55f4c398
  Tim Dettmers authored Jul 14, 2023
  
  55f4c398
- Changed CUDA setup to use PyTorch default; added a weak test. · 1ab6758b
  Tim Dettmers authored Jul 13, 2023
  
  1ab6758b
- Merge branch 'main' into bugfixes · ac155f74
  Tim Dettmers authored Jul 13, 2023
  
  ac155f74
- Merge pull request #375 from rapsealk/fix/libcuda-to-torch · e8df8d64
  Tim Dettmers authored Jul 13, 2023
```
Replace libcudart.so with PyTorch's CUDA APIs
```
  e8df8d64
- Fixed a bug in absmax float conversion. · c00402f1
  Tim Dettmers authored Jul 13, 2023
  
  c00402f1
- Merge pull request #567 from apbard/patch-1 · 6689afae
  Tim Dettmers authored Jul 13, 2023
```
[BugFix] replace view+continuous with reshape 
```
  6689afae
- Added documentation for NF4; failing 8-bit matmul; fixed absmax bug. #529 #543 · 67475257
  Tim Dettmers authored Jul 13, 2023
  
  67475257
- Added missing scipy requirement. Addressing #544 · 8a20cd86
  Tim Dettmers authored Jul 13, 2023
  
  8a20cd86
- Fixed bug caused by undefined default type of absmax. #553 · 097b1cc5
  Tim Dettmers authored Jul 13, 2023
  
  097b1cc5
- Added H100 support for CUDA 11.8 precompiled binaries. · 7b6cfe17
  Tim Dettmers authored Jul 13, 2023
  
  7b6cfe17
12 Jul, 2023 2 commits
- Bumped version after hotfix. · 817bdf63
  Tim Dettmers authored Jul 11, 2023
  
  817bdf63
- Fixed missing bias in bnb.matmul_4bit for inference; more tests. · 90b0ac57
  Tim Dettmers authored Jul 11, 2023
  
  90b0ac57
11 Jul, 2023 3 commits
- Test for bloom that fails with inference kernels. · dc96e9e7
  Tim Dettmers authored Jul 11, 2023
  
  dc96e9e7
- Bump version. · ae7cd6ad
  Tim Dettmers authored Jul 11, 2023
  
  ae7cd6ad
- Added more extensive gemv tests; blocksize guard for gemv. · ba51d95d
  Tim Dettmers authored Jul 11, 2023
  
  ba51d95d
10 Jul, 2023 13 commits
- Bump on version. · b8da4a16
  Tim Dettmers authored Jul 10, 2023
  
  b8da4a16
- Removed debugging statement. · a26a321e
  Tim Dettmers authored Jul 10, 2023
  
  a26a321e
- Fixed accidential deletion of limits in kernel. · 306f6b23
  Tim Dettmers authored Jul 10, 2023
  
  306f6b23
- Fixed potential memory leak. · 2221f4ce
  Tim Dettmers authored Jul 10, 2023
  
  2221f4ce
- Added generation tests. · 490153b2
  Tim Dettmers authored Jul 10, 2023
  
  490153b2
- Added ARCH guard for bfloat16 computations. · 1c774ece
  Tim Dettmers authored Jul 10, 2023
  
  1c774ece
- Fixed typo in cuda_install.sh. · 0a1cced3
  Tim Dettmers authored Jul 10, 2023
  
  0a1cced3
- Changelog and version bump. · 0d344b70
  Tim Dettmers authored Jul 10, 2023
  
  0d344b70
- Fixed Makefile and added CUDA 12.2 install. · 73aa4e0a
  Tim Dettmers authored Jul 10, 2023
  
  73aa4e0a
- Merge remote-tracking branch 'origin/inference' · 5f492d43
  Tim Dettmers authored Jul 10, 2023
  
  5f492d43
- Merge pull request #469 from shadeMe/linear-layer-device · 196d6f5d
  Tim Dettmers authored Jul 10, 2023
```
Add `device` parameter to `Linear` subclasses and `Embedding`
```
  196d6f5d
- Added fp32 compute type for gemv_4bit. · 5fab6734
  Tim Dettmers authored Jul 09, 2023
  
  5fab6734
- Added test for Param4bit.to() and fixed double quant behavior. · cef519c8
  Tim Dettmers authored Jul 09, 2023
  
  cef519c8
09 Jul, 2023 4 commits
- Fixed a bug where gemv_4bit would return a wrongly sized tensor. · 6a905be5
  Tim Dettmers authored Jul 09, 2023
  
  6a905be5
- Added double quantization support and tests. · 0f0390ac
  Tim Dettmers authored Jul 09, 2023
  
  0f0390ac
- Added FP4 fast inference support. · 94168d79
  Tim Dettmers authored Jul 09, 2023
  
  94168d79
- Added abitrary data types; fixed a bug for small matrices. · 4b88d69d
  Tim Dettmers authored Jul 09, 2023
  
  4b88d69d
08 Jul, 2023 2 commits
- Turning optimization (float accumulation). 185 vs 50. · eefbf602
  Tim Dettmers authored Jul 08, 2023
  
  eefbf602
- Added warp_shuffle indexing 185 vs 54. · 7e49b5b9
  Tim Dettmers authored Jul 08, 2023
  
  7e49b5b9
06 Jul, 2023 2 commits
- [BugFix] replace view+continuous with reshape · 463630dc
  Alessandro Pietro Bardelli authored Jul 06, 2023
  
  463630dc
- Merge branch 'main' into fix/libcuda-to-torch · a24aae30
  Jeongseok Kang authored Jul 06, 2023
  
  a24aae30
05 Jul, 2023 1 commit
- Added bfloat16 quantizations and tests. · 02fd80cb
  Tim Dettmers authored Jul 04, 2023
  
  02fd80cb
04 Jul, 2023 2 commits
- Vectorized loads, conflict free NF4; 52 vs 172. · dfe6900b
  Tim Dettmers authored Jul 04, 2023
  
  dfe6900b
- Initial 4-bit naive batch size 1, 81 vs 185. · f89ff93e
  Tim Dettmers authored Jul 03, 2023
  
  f89ff93e