github
/
ColossalAI
mirror of https://github.com/hpcaitech/ColossalAI
1
0
Fork 0
Code Issues Projects Releases Wiki Activity
3,757 Commits
32 Branches
48 Tags
232 MiB
Commit Graph

328 Commits (8e08c27e19d3f8dcfbae36dffcad0591c0cf9cfc)

Author SHA1 Message Date
duanjunwen c2fe3137e2
[hotfix] fix flash attn window_size err (#6132) 
* [fix] fix flash attn

* [hotfix] fix flash-atten version

* [fix] fix flash_atten version

* [fix] fix flash-atten versions

* [fix] fix flash-attn not enough values to unpack error

* [fix] fix test_ring_attn

* [fix] fix test ring attn
 2024-11-14 17:11:35 +08:00
Wenxuan Tan 62c13e7969
[Ring Attention] Improve comments (#6085) 
* improve comments

* improve comments

---------

Co-authored-by: Edenzzzz <wtan45@wisc.edu>
 2024-10-16 11:23:35 +08:00
Wang Binluo dcd41d0973
Merge pull request #6071 from wangbluo/ring_attention 
[Ring Attention] fix the 2d ring attn when using multiple machine
 2024-10-15 15:17:21 +08:00
wangbluo 83cf2f84fb fix 2024-10-15 14:50:27 +08:00
wangbluo bc7eeade33 fix 2024-10-15 13:28:33 +08:00
wangbluo fd92789af2 fix 2024-10-15 13:26:44 +08:00
wangbluo 6be9862aaf fix 2024-10-15 11:56:49 +08:00
wangbluo 3dc08c8a5a fix 2024-10-15 11:01:34 +08:00
wangbluo fe9208feac fix 2024-10-14 18:07:56 +08:00
wangbluo 3201377e94 fix 2024-10-14 18:06:24 +08:00
wangbluo 23199e34cc fix 2024-10-14 18:01:53 +08:00
wangbluo d891e50617 fix 2024-10-14 14:56:05 +08:00
wangbluo e1e86f9f1f fix 2024-10-14 11:45:35 +08:00
Tong Li 4c8e85ee0d
[Coati] Train DPO using PP (#6054) 
* update dpo

* remove unsupport plugin

* update msg

* update dpo

* remove unsupport plugin

* update msg

* update template

* update dataset

* add pp for dpo

* update dpo

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* add dpo fn

* update dpo

* update dpo

* update dpo

* update dpo

* minor update

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* update loss

* update help

* polish code

---------

Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
 2024-10-11 19:32:00 +08:00
wangbluo 1507a7528f fix 2024-10-11 06:20:34 +00:00
wangbluo 0002ae5956 fix 2024-10-11 14:16:21 +08:00
Hongxin Liu dc2cdaf3e8
[shardformer] optimize seq parallelism (#6086) 
* [shardformer] optimize seq parallelism

* [shardformer] fix gpt2 fused linear col

* [plugin] update gemini plugin

* [plugin] update moe hybrid plugin

* [test] update gpt2 fused linear test

* [shardformer] fix gpt2 fused linear reduce
 2024-10-11 13:44:40 +08:00
wangbluo efe3042bb2 fix 2024-10-10 18:38:47 +08:00
wangbluo 5ecc27e150 fix 2024-10-10 15:35:52 +08:00
wangbluo f98384aef6 fix 2024-10-10 15:17:06 +08:00
Hongxin Liu 646b3c5a90
[shardformer] fix linear 1d row and support uneven splits for fused qkv linear (#6084) 
* [tp] hotfix linear row

* [tp] support uneven split for fused linear

* [tp] support sp for fused linear

* [tp] fix gpt2 mlp policy

* [tp] fix gather fused and add fused linear row
 2024-10-10 14:34:45 +08:00
wangbluo b635dd0669 fix 2024-10-09 14:05:26 +08:00
wangbluo 3532f77b90 fix 2024-10-09 10:57:19 +08:00
wangbluo 3fab92166e fix 2024-09-26 18:03:09 +08:00
wangbluo 6705dad41b fix 2024-09-25 19:02:21 +08:00
wangbluo 91ed32c256 fix 2024-09-25 19:00:38 +08:00
wangbluo 6fb1322db1 fix 2024-09-25 18:56:18 +08:00
wangbluo 65c8297710 fix the attn 2024-09-25 18:51:03 +08:00
wangbluo cfd9eda628 fix the ring attn 2024-09-25 18:34:29 +08:00
botbw 4fa6b9509c
[moe] add parallel strategy for shared_expert && fix test for deepseek (#6063) 2024-09-18 10:09:01 +08:00
wangbluo 10e4f7da72 fix 2024-09-16 13:45:04 +08:00
Wang Binluo 37e35230ff
Merge pull request #6061 from wangbluo/sp_fix 
[sp] : fix the attention kernel for sp
 2024-09-14 20:54:35 +08:00
wangbluo 827ef3ee9a fix 2024-09-14 10:40:35 +00:00
Guangyao Zhang f20b066c59
[fp8] Disable all_gather intranode. Disable Redundant all_gather fp8 (#6059) 
* all_gather only internode, fix pytest

* fix cuda arch <89 compile pytest error

* fix pytest failure

* disable all_gather_into_tensor_flat_fp8

* fix fp8 format

* fix pytest

* fix conversations

* fix chunk tuple to list
 2024-09-14 10:40:01 +08:00
wangbluo b582319273 fix 2024-09-13 10:24:41 +00:00
wangbluo 0ad3129cb9 fix 2024-09-13 09:01:26 +00:00
wangbluo 0b14a5512e fix 2024-09-13 07:06:14 +00:00
botbw 696fced0d7
[fp8] fix missing fp8_comm flag in mixtral (#6057) 2024-09-13 14:30:05 +08:00
wangbluo dc032172c3 fix 2024-09-13 06:00:58 +00:00
wangbluo f393867cff fix 2024-09-13 05:24:52 +00:00
wangbluo 6eb8832366 fix 2024-09-13 05:06:56 +00:00
wangbluo 683179cefd fix 2024-09-13 03:40:56 +00:00
wangbluo 0a01e2a453 fix the attn 2024-09-13 03:38:35 +00:00
pre-commit-ci[bot] 216d54e374 [pre-commit.ci] auto fixes from pre-commit.com hooks 
for more information, see https://pre-commit.ci
 2024-09-13 02:38:40 +00:00
wangbluo fdd84b9087 fix the sp 2024-09-13 02:32:03 +00:00
botbw c54c4fcd15
[hotfix] moe hybrid parallelism benchmark & follow-up fix (#6048) 
* [example] pass use_fp8_comm flag to all plugins

* [example] add mixtral benchmark

* [moe] refine assertion and check

* [moe] fix mixtral & add more tests

* [moe] consider checking dp * sp group and moe_dp_group

* [mixtral] remove gate tp & add more tests

* [deepseek] fix tp & sp for deepseek

* [mixtral] minor fix

* [deepseek] add deepseek benchmark
 2024-09-10 17:30:53 +08:00
Wenxuan Tan 8fd25d6e09
[Feature] Split cross-entropy computation in SP (#5959) 
* halfway

* fix cross-PP-stage position id length diff bug

* fix typo

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* unified cross entropy func for all shardformer models

* remove redundant lines

* add basic ring attn; debug cross entropy

* fwd bwd logic complete

* fwd bwd logic complete; add experimental triton rescale

* precision tests passed

* precision tests passed

* fix typos and remove misc files

* update softmax_lse shape by new interface

* change tester name

* remove buffer clone; support packed seq layout

* add varlen tests

* fix typo

* all tests passed

* add dkv_group; fix mask

* remove debug statements

* adapt chatglm, command-R, qwen

* debug

* halfway

* fix cross-PP-stage position id length diff bug

* fix typo

* fix typo

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* unified cross entropy func for all shardformer models

* remove redundant lines

* add basic ring attn; debug cross entropy

* fwd bwd logic complete

* fwd bwd logic complete; add experimental triton rescale

* precision tests passed

* precision tests passed

* fix typos and remove misc files

* add sp_mode to benchmark; fix varlen interface

* update softmax_lse shape by new interface

* add varlen tests

* fix typo

* all tests passed

* add dkv_group; fix mask

* remove debug statements

* add comments

* q1 index only once

* remove events to simplify stream sync

* simplify forward/backward logic

* 2d ring forward passed

* 2d ring backward passed

* fixes

* fix ring attn loss

* 2D ring backward + llama passed

* merge

* update logger

* fix typo

* rebase

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* fix typo

* remove typos

* fixes

* support GPT

---------

Co-authored-by: Edenzzzz <wtan45@wisc.edu>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
 2024-09-10 12:06:50 +08:00
Hongxin Liu 17904cb5bf
Merge pull request #6012 from hpcaitech/feature/fp8_comm 
[fp8]  support fp8 communication and fp8 training for Colossalai
 2024-08-27 10:09:43 +08:00
wangbluo dae39999d7 fix 2024-08-26 03:45:42 +00:00
Wenxuan Tan 7cf9df07bc
[Hotfix] Fix llama fwd replacement bug (#6031) 
Co-authored-by: Edenzzzz <wtan45@wisc.edu>
 2024-08-23 15:44:27 +08:00