Commit Graph

461 Commits

Author SHA1 Message Date
Tim Dettmers
486488bccb Bumped version. 2023-07-14 12:55:57 -07:00
Tim Dettmers
6c6e5fcb53 Added changelog entry. 2023-07-14 12:55:04 -07:00
Tim Dettmers
55f4c398a0 Polished CUDA SETUP replacement and added docs. 2023-07-14 12:50:59 -07:00
Tim Dettmers
1ab6758b36 Changed CUDA setup to use PyTorch default; added a weak test. 2023-07-13 23:58:41 -07:00
Tim Dettmers
ac155f7415 Merge branch 'main' into bugfixes 2023-07-13 21:55:35 -07:00
Tim Dettmers
e8df8d64a2
Merge pull request #375 from rapsealk/fix/libcuda-to-torch
Replace libcudart.so with PyTorch's CUDA APIs
2023-07-13 21:54:47 -07:00
Tim Dettmers
c00402f17e Fixed a bug in absmax float conversion. 2023-07-13 21:47:38 -07:00
Tim Dettmers
6689afaec4
Merge pull request #567 from apbard/patch-1
[BugFix] replace view+continuous with reshape
2023-07-13 21:45:00 -07:00
Tim Dettmers
67475257a9 Added documentation for NF4; failing 8-bit matmul; fixed absmax bug. #529 #543 2023-07-13 21:41:43 -07:00
Tim Dettmers
8a20cd864b Added missing scipy requirement. Addressing #544 2023-07-13 21:25:07 -07:00
Tim Dettmers
097b1cc5da Fixed bug caused by undefined default type of absmax. #553 2023-07-13 21:23:33 -07:00
Tim Dettmers
7b6cfe1738 Added H100 support for CUDA 11.8 precompiled binaries. 2023-07-13 21:16:23 -07:00
Tim Dettmers
817bdf6325 Bumped version after hotfix. 2023-07-11 17:16:05 -07:00
Tim Dettmers
90b0ac57b0 Fixed missing bias in bnb.matmul_4bit for inference; more tests. 2023-07-11 17:13:33 -07:00
Tim Dettmers
dc96e9e7c8 Test for bloom that fails with inference kernels. 2023-07-11 15:40:20 -07:00
Tim Dettmers
ae7cd6ad14 Bump version. 2023-07-11 05:58:25 -07:00
Tim Dettmers
ba51d95d43 Added more extensive gemv tests; blocksize guard for gemv. 2023-07-11 05:55:49 -07:00
Tim Dettmers
b8da4a165a Bump on version. 2023-07-10 16:40:22 -07:00
Tim Dettmers
a26a321e07 Removed debugging statement. 2023-07-10 14:34:19 -07:00
Tim Dettmers
306f6b2362 Fixed accidential deletion of limits in kernel. 2023-07-10 14:24:33 -07:00
Tim Dettmers
2221f4cee0 Fixed potential memory leak. 2023-07-10 13:57:44 -07:00
Tim Dettmers
490153b29f Added generation tests. 2023-07-10 12:19:16 -07:00
Tim Dettmers
1c774ecebb Added ARCH guard for bfloat16 computations. 2023-07-10 09:53:23 -07:00
Tim Dettmers
0a1cced375 Fixed typo in cuda_install.sh. 2023-07-10 06:40:19 -07:00
Tim Dettmers
0d344b70ba Changelog and version bump. 2023-07-10 06:38:57 -07:00
Tim Dettmers
73aa4e0a33 Fixed Makefile and added CUDA 12.2 install. 2023-07-10 06:34:04 -07:00
Tim Dettmers
5f492d437e Merge remote-tracking branch 'origin/inference' 2023-07-10 06:24:24 -07:00
Tim Dettmers
196d6f5dc1
Merge pull request #469 from shadeMe/linear-layer-device
Add `device` parameter to `Linear` subclasses and `Embedding`
2023-07-10 06:17:13 -07:00
Tim Dettmers
5fab673442 Added fp32 compute type for gemv_4bit. 2023-07-09 21:06:01 -07:00
Tim Dettmers
cef519c89e Added test for Param4bit.to() and fixed double quant behavior. 2023-07-09 17:16:50 -07:00
Tim Dettmers
6a905be5ce Fixed a bug where gemv_4bit would return a wrongly sized tensor. 2023-07-09 15:34:02 -07:00
Tim Dettmers
0f0390acb2 Added double quantization support and tests. 2023-07-09 15:32:03 -07:00
Tim Dettmers
94168d79d7 Added FP4 fast inference support. 2023-07-09 14:46:19 -07:00
Tim Dettmers
4b88d69de7 Added abitrary data types; fixed a bug for small matrices. 2023-07-09 12:04:09 -07:00
Tim Dettmers
eefbf60270 Turning optimization (float accumulation). 185 vs 50. 2023-07-08 16:31:58 -07:00
Tim Dettmers
7e49b5b938 Added warp_shuffle indexing 185 vs 54. 2023-07-08 14:27:12 -07:00
Alessandro Pietro Bardelli
463630dc73
[BugFix] replace view+continuous with reshape 2023-07-06 12:26:03 +02:00
Jeongseok Kang
a24aae30bf Merge branch 'main' into fix/libcuda-to-torch 2023-07-06 15:43:42 +09:00
Tim Dettmers
02fd80cb81 Added bfloat16 quantizations and tests. 2023-07-04 19:58:31 -07:00
Tim Dettmers
dfe6900b94 Vectorized loads, conflict free NF4; 52 vs 172. 2023-07-04 15:20:10 -07:00
Tim Dettmers
f89ff93e26 Initial 4-bit naive batch size 1, 81 vs 185. 2023-07-03 18:45:38 -07:00
Tim Dettmers
4395d68cf6 Release 0.39.1. 2023-06-19 19:40:41 -07:00
Tim Dettmers
2d321a7524
Merge pull request #503 from TimDettmers/efficient_8bit_serialize
Make 8-bit serialization more memory-efficient (v2)
2023-06-19 11:28:30 -07:00
Max Ryabinin
b599fdb197 Only rearrange weight if it exists 2023-06-14 19:27:13 +02:00
Max Ryabinin
c1f3f56d2c Rearrange the weights directly in state dict before loading 2023-06-09 21:58:39 +02:00
Max Ryabinin
f734076e94 Improve memory efficiency of 8-bit serialization 2023-06-09 21:39:57 +02:00
Max Ryabinin
4fb37d45c1 Extract get_tile_inds to a separate function 2023-06-09 21:39:37 +02:00
shadeMe
db49ad43ab
Add device parameter to Embedding 2023-06-01 17:43:49 +02:00
shadeMe
9cac5dd1b6
Add device parameter to Linear subclasses 2023-06-01 17:43:30 +02:00
Tim Dettmers
e54d2730fc Added debugging functions. 2023-05-30 20:42:21 -07:00