Commit Graph

553 Commits

Author SHA1 Message Date
Tri Dao
e02fd588aa [Gen] Implement top-k and top-p sampling 2023-01-07 17:00:02 -08:00
Tri Dao
11be742aa3 [Gen] Test generation with rotary embedding 2023-01-07 14:37:54 -08:00
Tri Dao
8d9674ed08
Merge pull request #102 from Lamikins/main
fixed cross attention typeerror
2023-01-07 13:56:20 -08:00
Tri Dao
93383bd55b [TP] Implement TensorParallel without sequence parallel 2023-01-07 13:45:22 -08:00
Darius Lam
aec35fd67c fixed cross attention typeerror 2023-01-07 12:58:41 -08:00
Tri Dao
ce26d3d73d Bump to v0.2.7 2023-01-06 17:37:30 -08:00
Tri Dao
6738d9477d [LayerNorm] Implement RMS Norm 2023-01-06 17:34:22 -08:00
Tri Dao
a1f49a2b92 [Compilation] Change BOOL_SWITCH to fix Windows compilation
Follow xFormers's DISTPATCH_BOOL. Haven't tested it on Windows.
2023-01-06 14:40:58 -08:00
Tri Dao
a668890fcd [Gen] Add option to run generation with FT attention kernel 2023-01-03 22:10:31 -08:00
Tri Dao
be1afaa276 [Gen, FT] Use fp32 accum for FMA 2023-01-03 22:09:22 -08:00
Tri Dao
f266fc7262 [Gen, FT] Use tlength instead of params.timestep for rotary 2023-01-03 17:46:55 -08:00
Tri Dao
a01d1213d7 [Gen] Add kernel from FasterTransformer for benchmarking 2023-01-03 17:37:43 -08:00
Tri Dao
4cab4de5ea [TP] Put parallel embeddings in separate modules 2023-01-02 08:47:48 -08:00
Tri Dao
1ec09ebd90 [FusedDense] Limit matrix dims to 2M (instead of 64k) 2023-01-01 17:06:39 -08:00
Tri Dao
714c1b4f0f [Bert] Fix embedding layer norm before embedding dropout 2023-01-01 10:38:05 -08:00
Tri Dao
ef1ba918c6 [GPT] Refactor function to shard state_dict for TensorParallel 2023-01-01 00:09:33 -08:00
Tri Dao
65b4064b2a [FusedDense] Kick off input all_gather before weight dtype conversion 2022-12-31 22:47:34 -08:00
Tri Dao
71befc19e1 [Loss] Use flash_attn.losses.cross_entropy.CrossEntropyLoss 2022-12-31 22:43:28 -08:00
Tri Dao
cadfa396b8 [Docker] Set torchmetrics==0.10.3 2022-12-30 02:42:28 -08:00
Tri Dao
43798966cf [Docs] Fix formatting 2022-12-30 00:01:55 -08:00
Tri Dao
3c7cbfc195 [Docs] Mention that dropout_layer_norm supports all dims up to 6k 2022-12-29 23:55:33 -08:00
Tri Dao
85b8e3d334 [Docs] Mention that XPos's scale_base is recommended to be 512 2022-12-29 20:25:02 -08:00
Tri Dao
984d5204e2 Update training Dockerfile to use flash-attn==0.2.6 2022-12-29 15:12:33 -08:00
Tri Dao
029617179f
Merge pull request #95 from Quentin-Anthony/patch-1
Add gpt-neox adoption
2022-12-28 15:36:36 -08:00
Quentin Anthony
d2a69a55e2
Add gpt-neox adoption 2022-12-28 18:33:58 -05:00
Tri Dao
a6ec1782dc Bump to v0.2.6 2022-12-27 22:05:20 -08:00
Tri Dao
63670fd84a Implement generation for GPT 2022-12-27 21:01:50 -08:00
Tri Dao
9d797d8848 Support loading GPT2 weights from Huggingface 2022-12-27 11:22:48 -08:00
Tri Dao
c6ecd40a59 Tweak CrossEntropyLoss to take process_group in init 2022-12-27 10:47:43 -08:00
Caleb Thomas
c9a649805b Add a simple tutorial to README.md 2022-12-27 14:13:59 +08:00
Tri Dao
b4018a5028 Implement Tensor Parallel for GPT model 2022-12-26 16:22:43 -08:00
Tri Dao
78225c5366 Implement Tensor Parallel for GPT2Embeddings 2022-12-25 14:29:53 -08:00
Tri Dao
a8cfe51551 Implement Tensor Parallel for transformer Block 2022-12-25 14:08:21 -08:00
Tri Dao
1e712ea8b0 Implement TensorParallel for MHA 2022-12-25 11:39:55 -08:00
Tri Dao
226a1b721d Implement TensorParallel for FusedDense and FusedDenseGeluDense 2022-12-24 11:48:56 -08:00
Tri Dao
dff68c2b22 Add smoothing for CrossEntropyParallel, rename to CrossEntropyLoss 2022-12-23 14:51:08 -08:00
Tri Dao
e68ebbe89a Simplify FusedDense 2022-12-22 21:25:31 -08:00
Tri Dao
1bc6e5b09c Bump to v0.2.5 2022-12-21 14:33:18 -08:00
Tri Dao
496e4f528c Implement XPos (Sun et al.) 2022-12-21 14:17:58 -08:00
Tri Dao
c2407dec96 Fix typo in config: train.gpu -> train.gpu_mem 2022-12-21 13:42:30 -08:00
Tri Dao
13cdceb377 Implement last_layer_subset optimization for BERT 2022-12-19 22:18:46 -08:00
Tri Dao
5fb6df0e04 Implement BERT 2022-12-18 21:47:27 -08:00
Tri Dao
dc24c22603
Merge pull request #92 from ploshkin/rm-shape-asserts
Fix slicing dimensions in rotary embeddings
2022-12-17 11:22:06 -08:00
Alexander Ploshkin
ee8984d2be add asserts for sin shape 2022-12-17 13:34:57 +04:00
Alexander Ploshkin
c7c66976cc fix slicing dimensions 2022-12-16 15:39:06 +04:00
Tri Dao
b78f5a392d [Docs] Mention Megatron-LM 2022-12-15 19:49:04 -08:00
Tri Dao
ece8f05d09 [Docs] Mention PubMedGPT 2022-12-15 19:44:59 -08:00
Alexander Ploshkin
96656b9323 Remove redundant shape asserts in rotary embeddings 2022-12-15 18:13:21 +04:00
Tri Dao
04c4c6106e Bump to v0.2.4 2022-12-14 14:49:26 -08:00
Tri Dao
6b5f271c6d [Triton] Avoid einops repeat by using Tensor.expand 2022-12-14 14:48:41 -08:00