Commit Graph

2111 Commits (591a13bf7e39c18dbe1f49252047b2f6b73408d4)

Author SHA1 Message Date
duanjunwen 591a13bf7e [fix] fix optim bwd; 2024-09-02 11:19:42 +00:00
duanjunwen 77fe44286c [fix] rm zbv in hybridplugin 2024-09-02 10:00:43 +00:00
duanjunwen 6d18d38d5c [feat] update test; rm comments; 2024-09-02 09:50:47 +00:00
duanjunwen a7b767b071 [fix] fix communication_map; 2024-08-30 05:56:02 +00:00
duanjunwen 8eb6eac225 [fix] fix optim bwd; add license for v_schedule; remove redundant attributes; fix schedule loop "while"--> "for"; add communication dict; 2024-08-30 05:42:43 +00:00
duanjunwen 6af81d8c0d [feat] add fwd_bwd_step, run_fwd_only; 2024-08-30 02:47:52 +00:00
duanjunwen 48ba22dbfd [feat] fix optimizer bwd b & w; support return accum loss & output 2024-08-29 08:54:45 +00:00
duanjunwen 4c4b01b859 [feat] add optim backward_b_by_grad 2024-08-29 03:16:59 +00:00
duanjunwen fe209164f1 [feat] add apply v_schedule graph; p & p.grad assert err exist; 2024-08-27 10:29:39 +00:00
duanjunwen 8b37323f16 [feat] add run_fwd_bwd_with_microbatch (replace input) & test; add p&p.grad assert close test & all pass; 2024-08-27 09:31:38 +00:00
duanjunwen 9e0bd1af00 [fix] fix ci test; add pytest; 2024-08-27 08:00:23 +00:00
duanjunwen 283c9ff5d2 [fix] rm useless assign and comments; 2024-08-27 07:31:58 +00:00
duanjunwen 1b4bb2beeb [feat] add comments for ZBV func; 2024-08-27 07:11:50 +00:00
duanjunwen 5e09c8b4e1 [feat] split communication and calculation; fix pop empty send_bwd_buffer error; 2024-08-27 06:29:13 +00:00
duanjunwen 1d75045c37 [feat] add test run_fwd_bwd automatic scheduling; 2024-08-26 11:21:56 +00:00
duanjunwen fd5526b76e Merge branch 'main' into dev/zero_bubble 2024-08-26 04:03:20 +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
duanjunwen c18ef060cf [feat] add dw test; 2024-08-23 06:04:12 +00:00
duanjunwen ee9baedadf [feat] add zerobubble pp (just a frame now); add POC test for dx_dw; add test for zerobubble; 2024-08-22 10:25:34 +00:00
Hongxin Liu 0d3b0bd864
[plugin] add cast inputs option for zero (#6003) (#6022) 2024-08-21 10:21:26 +08:00
Edenzzzz dcc44aab8d
[misc] Use dist logger in plugins (#6011)
* use dist logger in plugins

* remove trash

* print on rank 0

---------

Co-authored-by: Edenzzzz <wtan45@wisc.edu>
2024-08-20 10:32:41 +08:00
Edenzzzz f1c3266a94
overlap kv comm with output rescale (#6017)
Co-authored-by: Edenzzzz <wtan45@wisc.edu>
2024-08-19 14:08:17 +08:00
Hongxin Liu 26493b97d3
[misc] update compatibility (#6008)
* [misc] update compatibility

* [misc] update requirements

* [devops] disable requirements cache

* [test] fix torch ddp test

* [test] fix rerun on address in use

* [test] fix lazy init
2024-08-16 18:49:14 +08:00
Edenzzzz f5c84af0b0
[Feature] Zigzag Ring attention (#5905)
* 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

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

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

* add sp_mode to benchmark; fix varlen interface

* 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

---------

Co-authored-by: Edenzzzz <wtan45@wisc.edu>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
2024-08-16 13:56:38 +08:00
Haze188 887d2d579b
[misc] Bypass the huggingface bug to solve the mask mismatch problem (#5991) 2024-08-15 14:40:26 +08:00
Hongxin Liu 406f984063
[plugin] add cast inputs option for zero (#6003) 2024-08-15 10:41:22 +08:00
Tong Li ceb1e262e7
fix sync condition (#6000) 2024-08-14 11:22:39 +08:00
Edenzzzz b4d2377d4c
[Hotfix] Avoid fused RMSnorm import error without apex (#5985)
Co-authored-by: Edenzzzz <wtan45@wisc.edu>
2024-08-09 18:17:09 +08:00
Wang Binluo 75c963686f
[lora] lora support hybrid parallel plugin (#5956)
* lora support hybrid plugin

* fix

* fix

* fix

* fix
2024-08-02 10:36:58 +08:00
botbw 62cdac6b7b [chore] remove redundant test case, print string & reduce test tokens 2024-08-01 10:06:59 +08:00
botbw d1d1ab871e [moe] solve dp axis issue 2024-08-01 10:06:59 +08:00
botbw 65daa87627 [doc] add MoeHybridParallelPlugin docstring 2024-08-01 10:06:59 +08:00
hxwang 7bedd03739 [moe] remove force_overlap_comm flag and add warning instead 2024-08-01 10:06:59 +08:00
hxwang f7c5485ed6 [chore] docstring 2024-08-01 10:06:59 +08:00
haze188 7e737df5ad [misc] remove useless condition 2024-08-01 10:06:59 +08:00
haze188 70793ce9ed [misc] fix ci failure: change default value to false in moe plugin 2024-08-01 10:06:59 +08:00
hxwang 606b0891ed [chore] change moe_pg_mesh to private 2024-08-01 10:06:59 +08:00
hxwang 5b4c12381b Revert "[moe] implement submesh initialization"
This reverts commit 2f9bce6686.
2024-08-01 10:06:59 +08:00
hxwang cb01c0d5ce [moe] refactor mesh assignment 2024-08-01 10:06:59 +08:00
haze188 034020bd04 [misc] remove debug/print code 2024-08-01 10:06:59 +08:00
hxwang c3dc9b4dba [deepseek] replace attn (a workaround for bug in transformers) 2024-08-01 10:06:59 +08:00
hxwang 6c39f0b144 [test] add check 2024-08-01 10:06:59 +08:00
haze188 b2952a5982 [moe] deepseek moe sp support 2024-08-01 10:06:59 +08:00
botbw 96d0fbc531 [bug] fix: somehow logger hangs the program 2024-08-01 10:06:59 +08:00
hxwang 067e18f7e9 [test] fix test: test_zero1_2 2024-08-01 10:06:59 +08:00
hxwang 74b03de3f9 [moe] remove ops 2024-08-01 10:06:59 +08:00
hxwang 70c9924d0d [chore] solve moe ckpt test failure and some other arg pass failure 2024-08-01 10:06:59 +08:00
hxwang 46037c2ccd [chore] minor fix after rebase 2024-08-01 10:06:59 +08:00
hxwang 803878b2fd [moe] full test for deepseek and mixtral (pp + sp to fix) 2024-08-01 10:06:59 +08:00
hxwang 7077d38d5a [moe] finalize test (no pp) 2024-08-01 10:06:59 +08:00