Commits · 1ec09ebd90c28b936538d193f275c99f06d5e8d1 · gaoqiong / flash-attention

02 Jan, 2023 1 commit
- [FusedDense] Limit matrix dims to 2M (instead of 64k) · 1ec09ebd
  Tri Dao authored Jan 01, 2023
  
  1ec09ebd
01 Jan, 2023 4 commits
- [Bert] Fix embedding layer norm before embedding dropout · 714c1b4f
  Tri Dao authored Jan 01, 2023
  
  714c1b4f
- [GPT] Refactor function to shard state_dict for TensorParallel · ef1ba918
  Tri Dao authored Jan 01, 2023
  
  ef1ba918
- [FusedDense] Kick off input all_gather before weight dtype conversion · 65b4064b
  Tri Dao authored Dec 31, 2022
  
  65b4064b
- [Loss] Use flash_attn.losses.cross_entropy.CrossEntropyLoss · 71befc19
  Tri Dao authored Dec 31, 2022
  
  71befc19
30 Dec, 2022 4 commits
- [Docker] Set torchmetrics==0.10.3 · cadfa396
  Tri Dao authored Dec 30, 2022
  
  cadfa396
- [Docs] Fix formatting · 43798966
  Tri Dao authored Dec 30, 2022
  
  43798966
- [Docs] Mention that dropout_layer_norm supports all dims up to 6k · 3c7cbfc1
  Tri Dao authored Dec 29, 2022
  
  3c7cbfc1
- [Docs] Mention that XPos's scale_base is recommended to be 512 · 85b8e3d3
  Tri Dao authored Dec 29, 2022
  
  85b8e3d3
29 Dec, 2022 1 commit
- Update training Dockerfile to use flash-attn==0.2.6 · 984d5204
  Tri Dao authored Dec 29, 2022
  
  984d5204
28 Dec, 2022 4 commits
- Merge pull request #95 from Quentin-Anthony/patch-1 · 02961717
  Tri Dao authored Dec 28, 2022
```
Add gpt-neox adoption
```
  02961717
- Add gpt-neox adoption · d2a69a55
  Quentin Anthony authored Dec 28, 2022
  
  d2a69a55
- Bump to v0.2.6 · a6ec1782
  Tri Dao authored Dec 27, 2022
  
  a6ec1782
- Implement generation for GPT · 63670fd8
  Tri Dao authored Dec 27, 2022
  
  63670fd8
27 Dec, 2022 3 commits
- Support loading GPT2 weights from Huggingface · 9d797d88
  Tri Dao authored Dec 27, 2022
  
  9d797d88
- Tweak CrossEntropyLoss to take process_group in init · c6ecd40a
  Tri Dao authored Dec 27, 2022
  
  c6ecd40a
- Implement Tensor Parallel for GPT model · b4018a50
  Tri Dao authored Dec 25, 2022
  
  b4018a50
25 Dec, 2022 3 commits
- Implement Tensor Parallel for GPT2Embeddings · 78225c53
  Tri Dao authored Dec 25, 2022
  
  78225c53
- Implement Tensor Parallel for transformer Block · a8cfe515
  Tri Dao authored Dec 25, 2022
  
  a8cfe515
- Implement TensorParallel for MHA · 1e712ea8
  Tri Dao authored Dec 24, 2022
  
  1e712ea8
24 Dec, 2022 1 commit
- Implement TensorParallel for FusedDense and FusedDenseGeluDense · 226a1b72
  Tri Dao authored Dec 23, 2022
  
  226a1b72
23 Dec, 2022 2 commits
- Add smoothing for CrossEntropyParallel, rename to CrossEntropyLoss · dff68c2b
  Tri Dao authored Dec 23, 2022
  
  dff68c2b
- Simplify FusedDense · e68ebbe8
  Tri Dao authored Dec 22, 2022
  
  e68ebbe8
21 Dec, 2022 3 commits
- Bump to v0.2.5 · 1bc6e5b0
  Tri Dao authored Dec 21, 2022
  
  1bc6e5b0
- Implement XPos (Sun et al.) · 496e4f52
  Tri Dao authored Dec 21, 2022
  
  496e4f52
- Fix typo in config: train.gpu -> train.gpu_mem · c2407dec
  Tri Dao authored Dec 21, 2022
  
  c2407dec
20 Dec, 2022 1 commit
- Implement last_layer_subset optimization for BERT · 13cdceb3
  Tri Dao authored Dec 19, 2022
  
  13cdceb3
19 Dec, 2022 1 commit
- Implement BERT · 5fb6df0e
  Tri Dao authored Dec 18, 2022
  
  5fb6df0e
17 Dec, 2022 2 commits
- Merge pull request #92 from ploshkin/rm-shape-asserts · dc24c226
  Tri Dao authored Dec 17, 2022
```
Fix slicing dimensions in rotary embeddings
```
  dc24c226
- add asserts for sin shape · ee8984d2
  Alexander Ploshkin authored Dec 17, 2022
  
  ee8984d2
16 Dec, 2022 3 commits
- fix slicing dimensions · c7c66976
  Alexander Ploshkin authored Dec 16, 2022
  
  c7c66976
- [Docs] Mention Megatron-LM · b78f5a39
  Tri Dao authored Dec 15, 2022
  
  b78f5a39
- [Docs] Mention PubMedGPT · ece8f05d
  Tri Dao authored Dec 15, 2022
  
  ece8f05d
15 Dec, 2022 1 commit
- Remove redundant shape asserts in rotary embeddings · 96656b93
  Alexander Ploshkin authored Dec 15, 2022
  
  96656b93
14 Dec, 2022 2 commits
- Bump to v0.2.4 · 04c4c610
  Tri Dao authored Dec 14, 2022
  
  04c4c610
- [Triton] Avoid einops repeat by using Tensor.expand · 6b5f271c
  Tri Dao authored Dec 14, 2022
  
  6b5f271c
13 Dec, 2022 3 commits
- Fix the case when dout is not contiguous · 88c4e5db
  Tri Dao authored Dec 13, 2022
  
  88c4e5db
- Bump to v0.2.3 · a1a5d2ee
  Tri Dao authored Dec 13, 2022
  
  a1a5d2ee
- [LayerNorm] Support taking subset of input or subset of output · 5db33051
  Tri Dao authored Dec 12, 2022
  
  5db33051
11 Dec, 2022 1 commit
- [LayerNorm] Fuse LayerScale · ae137ed1
  Tri Dao authored Dec 10, 2022
  
  ae137ed1