Commit Graph

133 Commits

Author SHA1 Message Date
Tri Dao
85b51d61ee Bump version to 1.0.7 2023-05-30 14:18:44 -07:00
Tri Dao
48bc6eacd6 [Gen] Add rotary base as an argument to FT attention kernel 2023-05-30 13:38:34 -07:00
Kirthi Shankar Sivamani
dd9c3a1fc2 bump to v1.0.6
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>
2023-05-26 17:44:10 -07:00
Max H. Gerlach
31f78a9814 Allow adding an optional local version to the package version 2023-05-19 17:27:41 +02:00
Federico Berto
69f5f7d0a2 [BugFix] cannot unpack non-iterable NoneType object 2023-05-07 03:07:44 +09:00
Federico Berto
3889ba168b [BugFix] cannot unpack non-iterable NoneType object 2023-05-07 03:07:30 +09:00
Tri Dao
a9a4b4e4f2 [LLaMa] Fix last norm layer to use RMSNorm instead of LayerNorm 2023-05-04 23:39:43 -07:00
Tri Dao
fcab93b43a [Gen] Minor tweak to allocate_inference_cache 2023-04-21 11:56:47 -07:00
Tri Dao
ba2fe7f378 [Gen] Move allocate_inference_cache to within the model 2023-04-20 18:15:12 -07:00
Tri Dao
3da42d24b1 [GPT] Add option to only return the logit for the last token 2023-04-20 17:21:08 -07:00
Tri Dao
311d6606bf [Gen] Fix FT kernel smem size, CG when batch size changed 2023-04-20 17:03:13 -07:00
Tri Dao
96d10f6545 Implement LLaMa 2023-04-18 21:51:35 -07:00
Tri Dao
b630aef53f Implement GatedMlp 2023-04-18 03:37:14 -07:00
Tri Dao
ac3b684cdb Have a separate nn.Dropout module in SelfAttention module 2023-04-17 22:34:05 -07:00
Kirthi Shankar Sivamani
a0997bc77c
Merge branch 'HazyResearch:main' into enable_cuda_graph_capture 2023-04-14 21:45:37 -07:00
Tri Dao
605655bc66 [Gen] Fix FT kernel when using CG 2023-04-14 16:50:01 -07:00
Kirthi Shankar Sivamani
081c2b012a
Merge branch 'HazyResearch:main' into enable_cuda_graph_capture 2023-04-13 19:36:45 -07:00
Tri Dao
1c9ef9b399 [Gen] Measure prompt processing + decoding time, not just decoding 2023-04-13 15:39:56 -07:00
Tri Dao
6f6e9a9aaf [FusedDense] Enable sqrelu activation in FusedMLP 2023-04-13 15:29:32 -07:00
Kirthi Shankar Sivamani
7d25a4ec4f Handle FlashAttnQKVPackedSplitFunc by making rng_state optional in backward
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>
2023-04-13 06:25:52 +00:00
Kirthi Shankar Sivamani
315fd31f0c
Merge branch 'HazyResearch:main' into enable_cuda_graph_capture 2023-04-12 22:42:24 -07:00
Zhiyuan Chen
8c42415664
make mlp hidden_features defaults to 4*in_features 2023-04-13 11:08:21 +08:00
Kirthi Shankar Sivamani
31018c5fa0 Support CUDA graph capture
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>
2023-04-12 16:53:22 -07:00
Tri Dao
d6fc860573
Merge pull request #147 from ksivaman/add_deterministic_execution_option
Add option for deterministic execution
2023-03-31 17:32:50 -04:00
Tri Dao
393882bc08 [LayerNorm] Implement LN with parallel residual, support dim 8k 2023-03-31 14:23:45 -07:00
Kirthi Shankar Sivamani
b6aa059bbf Add option for deterministic execution 2023-03-30 18:23:35 -07:00
Tri Dao
993d12448e Implement GPT-NeoX 2023-03-29 01:21:25 -07:00
Tri Dao
f5d0fbd468 [FT] Fix FT's single query attention for bf16 hdim128 rotary 2023-03-28 21:27:00 -07:00
Tri Dao
4d87e4d875 Implement GPT-J 2023-03-22 16:16:58 -07:00
Tri Dao
5d079fdd7a [Triton] Fix benchmark_causal, mention Triton version 2023-03-22 00:51:16 -07:00
Tri Dao
dc08ea1c33 Support H100 for other CUDA extensions 2023-03-15 16:59:27 -07:00
Vik Paruchuri
3165398074 Remove unused kwargs in flashattention 2023-03-15 10:36:19 -07:00
Tri Dao
e45a46a5b7 [Rotary] Implement GPT-J style (interleaved) rotary 2023-03-14 14:35:53 -07:00
Tri Dao
78b7a1dc18 [OPT] Load fp16 weights on CPU before moving to GPU 2023-01-22 17:01:32 -08:00
Tri Dao
eb33e587e9 [LayerNorm] Rename x1 -> residual 2023-01-19 13:07:27 -08:00
Tri Dao
f68d41ec77 [Gen] Add OPT to generation test 2023-01-17 19:59:06 -08:00
Tri Dao
88173a1aaf [FusedDense] Support relu, rename FusedDenseGeluDense -> FusedMLP 2023-01-17 18:12:27 -08:00
Tri Dao
780e8eeabb [ViT] Support timm checkpoint, add tests 2023-01-16 01:20:34 -08:00
Tri Dao
2ec7d3f72c
Merge pull request #105 from jamaliki/patch-1
Change default dropout value in documentation
2023-01-15 23:01:20 -08:00
Tri Dao
ef085cfcda [ViT] Fix extra norm_0, use new LN order in Block 2023-01-15 22:58:56 -08:00
Tri Dao
ff34123bd4 Reorder LN in Block, support OPT 2023-01-15 22:14:31 -08:00
Tri Dao
7c2191542a [Gen] Make generation work with Tensor Parallel 2023-01-15 11:34:27 -08:00
Kiarash Jamali
41cb909741
Change default dropout value in documentation
Documentation says default is 0.1, but the code has attention_dropout default at 0.0
2023-01-13 10:50:07 +00:00
Tri Dao
f95c2fc108 [Gen] Remove commented code 2023-01-07 19:06:39 -08:00
Tri Dao
b48599002a [Gen] Add timing option 2023-01-07 19:05:09 -08:00
Tri Dao
0938298e4c [Gen] Adjust shape of kv_cache when using FT 2023-01-07 17:27:54 -08:00
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