Commit Graph

  • 1a5847e6d1 fix the merge wangbluo 2024-08-19 03:28:29 +00:00
  • 52289e4c63 Merge branch 'fp8_merge' of https://github.com/wangbluo/ColossalAI into fp8_merge wangbluo 2024-08-19 02:27:30 +00:00
  • 02636c5bef fix the merge wangbluo 2024-08-19 02:26:52 +00:00
  • 81272e9d00 [pre-commit.ci] auto fixes from pre-commit.com hooks pre-commit-ci[bot] 2024-08-17 09:37:37 +00:00
  • 4cf79fa275 merge wangbluo 2024-08-17 09:34:18 +00:00
  • 26493b97d3 [misc] update compatibility (#6008) Hongxin Liu 2024-08-16 18:49:14 +08:00
  • f5c84af0b0 [Feature] Zigzag Ring attention (#5905) Edenzzzz 2024-08-16 13:56:38 +08:00
  • 0a51319113 [fp8] zero support fp8 linear. (#6006) flybird11111 2024-08-16 10:13:07 +08:00
  • 3f09a6145f [fp8] add use_fp8 option for MoeHybridParallelPlugin (#6009) Wang Binluo 2024-08-16 10:12:50 +08:00
  • 20722a8c93 [fp8]update reduce-scatter test (#6002) flybird11111 2024-08-15 14:40:54 +08:00
  • 887d2d579b [misc] Bypass the huggingface bug to solve the mask mismatch problem (#5991) Haze188 2024-08-15 14:40:26 +08:00
  • 4dd03999ec [pre-commit.ci] pre-commit autoupdate (#5995) pre-commit-ci[bot] 2024-08-15 14:40:03 +08:00
  • 1a2e90dcc1 [fp8] linear perf enhancement botbw 2024-08-15 03:12:08 +00:00
  • 406f984063 [plugin] add cast inputs option for zero (#6003) Hongxin Liu 2024-08-15 10:41:22 +08:00
  • 88fa096d78 [fp8] update torch.compile for linear_fp8 to >= 2.4.0 (#6004) botbw 2024-08-15 10:14:42 +08:00
  • 597b206001 [fp8] support asynchronous FP8 communication (#5997) flybird11111 2024-08-14 14:08:19 +08:00
  • ceb1e262e7 fix sync condition (#6000) Tong Li 2024-08-14 11:22:39 +08:00
  • 0978080a69 [fp8] refactor fp8 linear with compile (#5993) Hongxin Liu 2024-08-13 16:07:26 +08:00
  • b2483c8e31 [fp8] support hybrid parallel plugin (#5982) Wang Binluo 2024-08-12 18:17:05 +08:00
  • ed97d3a5d3 [Chat] fix readme (#5989) YeAnbang 2024-08-12 14:55:17 +08:00
  • f7b4fb5f07 [pre-commit.ci] auto fixes from pre-commit.com hooks ColossalChat pre-commit-ci[bot] 2024-08-12 03:12:22 +00:00
  • 2d642eea0f fix readme, tokenization fully tested root 2024-08-12 11:08:01 +08:00
  • 0f70c3440d fix readme, tokenization fully tested root 2024-08-12 11:07:09 +08:00
  • dccc9e14ba fix readme root 2024-08-12 11:01:44 +08:00
  • f1a3a326c4 [fp8]Moe support fp8 communication (#5977) flybird11111 2024-08-09 18:26:02 +08:00
  • b4d2377d4c [Hotfix] Avoid fused RMSnorm import error without apex (#5985) Edenzzzz 2024-08-09 18:17:09 +08:00
  • e4aadeee20 [fp8] use torch compile (torch >= 2.3.0) (#5979) botbw 2024-08-09 15:51:06 +08:00
  • 8241c0c054 [fp8] support gemini plugin (#5978) Hongxin Liu 2024-08-09 14:09:48 +08:00
  • ad3fa4f49c [Hotfix] README link (#5966) Tong Li 2024-08-08 18:04:47 +08:00
  • 4b9bec8176 [test ci]Feature/fp8 comm (#5981) flybird11111 2024-08-08 17:19:21 +08:00
  • b480eec738 [Feature]: support FP8 communication in DDP, FSDP, Gemini (#5928) Hanks 2024-08-08 15:55:01 +08:00
  • 7739629b9d fix (#5976) flybird11111 2024-08-07 18:58:39 +08:00
  • ccabcf6485 [fp8] support fp8 amp for hybrid parallel plugin (#5975) Hongxin Liu 2024-08-07 18:21:08 +08:00
  • 76ea16466f [fp8] add fp8 linear (#5967) Hongxin Liu 2024-08-07 15:41:49 +08:00
  • 9179d4088e [Docs] clarify launch port Edenzzzz 2024-08-07 13:53:48 +08:00
  • afb26de873 [fp8]support all2all fp8 (#5953) flybird11111 2024-08-06 16:58:23 +08:00
  • 0c10afd372 [FP8] rebase main (#5963) flybird11111 2024-08-06 16:29:37 +08:00
  • 53cb9606bd [Feature] llama shardformer fp8 support (#5938) Guangyao Zhang 2024-08-05 10:05:47 +08:00
  • c297e21bea Merge pull request #5961 from ver217/feature/zeor-fp8 Hanks 2024-08-02 20:38:58 +08:00
  • fe71917851 Merge pull request #5962 from hpcaitech/colossalchat YeAnbang 2024-08-02 17:32:41 +08:00
  • 0b2d55c4ab Support overall loss, update KTO logging colossalchat YeAnbang 2024-08-02 06:51:38 +00:00
  • 91e596d017 [test] add zero fp8 test case ver217 2024-08-02 11:28:38 +08:00
  • ae486ce005 [fp8] add fp8 comm for low level zero ver217 2024-08-02 11:12:12 +08:00
  • 75c963686f [lora] lora support hybrid parallel plugin (#5956) Wang Binluo 2024-08-02 10:36:58 +08:00
  • 19d1510ea2 [feat] Dist Loader for Eval (#5950) Tong Li 2024-08-02 10:06:25 +08:00
  • 62cdac6b7b [chore] remove redundant test case, print string & reduce test tokens botbw 2024-07-30 02:56:39 +00:00
  • d1d1ab871e [moe] solve dp axis issue botbw 2024-07-26 07:32:19 +00:00
  • 65daa87627 [doc] add MoeHybridParallelPlugin docstring botbw 2024-07-26 06:05:01 +00:00
  • 7bedd03739 [moe] remove force_overlap_comm flag and add warning instead hxwang 2024-07-25 09:49:57 +00:00
  • f7c5485ed6 [chore] docstring hxwang 2024-07-25 09:07:49 +00:00
  • 7e737df5ad [misc] remove useless condition haze188 2024-07-25 08:40:53 +00:00
  • 70793ce9ed [misc] fix ci failure: change default value to false in moe plugin haze188 2024-07-25 07:50:41 +00:00
  • 12d043ca00 [misc] remove incompatible test config haze188 2024-07-25 06:40:54 +00:00
  • 606b0891ed [chore] change moe_pg_mesh to private hxwang 2024-07-25 06:34:22 +00:00
  • 5b4c12381b Revert "[moe] implement submesh initialization" hxwang 2024-07-25 06:32:02 +00:00
  • cb01c0d5ce [moe] refactor mesh assignment hxwang 2024-07-25 06:19:54 +00:00
  • 034020bd04 [misc] remove debug/print code haze188 2024-07-24 03:10:57 +00:00
  • 59bcf56c60 [misc] skip redunant test haze188 2024-07-24 02:59:35 +00:00
  • c3dc9b4dba [deepseek] replace attn (a workaround for bug in transformers) hxwang 2024-07-23 12:56:58 +00:00
  • 6c39f0b144 [test] add check hxwang 2024-07-23 09:38:05 +00:00
  • b2952a5982 [moe] deepseek moe sp support haze188 2024-07-23 06:39:49 +00:00
  • 96d0fbc531 [bug] fix: somehow logger hangs the program botbw 2024-07-23 06:17:51 +00:00
  • 067e18f7e9 [test] fix test: test_zero1_2 hxwang 2024-07-22 05:36:20 +00:00
  • 74b03de3f9 [moe] remove ops hxwang 2024-07-22 04:00:42 +00:00
  • 70c9924d0d [chore] solve moe ckpt test failure and some other arg pass failure hxwang 2024-07-22 03:40:34 +00:00
  • 52d346f2a5 [pre-commit.ci] auto fixes from pre-commit.com hooks pre-commit-ci[bot] 2024-07-19 07:54:40 +00:00
  • 46037c2ccd [chore] minor fix after rebase hxwang 2024-07-19 07:53:40 +00:00
  • 803878b2fd [moe] full test for deepseek and mixtral (pp + sp to fix) hxwang 2024-07-19 06:11:11 +00:00
  • 7077d38d5a [moe] finalize test (no pp) hxwang 2024-07-18 13:36:18 +00:00
  • 2cddeac717 moe sp + ep bug fix haze188 2024-07-18 10:08:06 +00:00
  • 877d94bb8c [moe] init moe plugin comm setting with sp hxwang 2024-07-18 08:37:06 +00:00
  • 09d6280d3e [chore] minor fix hxwang 2024-07-18 03:53:51 +00:00
  • 404b16faf3 [Feature] MoE Ulysses Support (#5918) Haze188 2024-07-18 11:37:56 +08:00
  • 3e2b6132b7 [moe] clean legacy code hxwang 2024-07-16 09:08:31 +00:00
  • 74eccac0db [moe] test deepseek hxwang 2024-07-16 10:10:40 +00:00
  • dc583aa576 [moe] implement tp botbw 2024-07-16 06:03:57 +00:00
  • 0b5bbe9ce4 [test] add mixtral modelling test botbw 2024-07-15 06:43:27 +00:00
  • 102b784a10 [chore] arg pass & remove drop token hxwang 2024-07-12 09:08:16 +00:00
  • 8dbb86899d [chore] trivial fix botbw 2024-07-12 07:04:17 +00:00
  • 014faf6c5a [chore] manually revert unintended commit botbw 2024-07-12 03:29:16 +00:00
  • 9b9b76bdcd [moe] add mixtral dp grad scaling when not all experts are activated botbw 2024-07-12 03:27:20 +00:00
  • e28e05345b [moe] implement submesh initialization botbw 2024-07-11 05:50:20 +00:00
  • 5ed5e8cfba solve hang when parallel mode = pp + dp haze188 2024-07-11 02:12:44 +00:00
  • fe24789eb1 [misc] solve booster hang by rename the variable haze188 2024-07-09 09:44:04 +00:00
  • 13b48ac0aa [zero] solve hang botbw 2024-07-09 08:14:00 +00:00
  • b5bfeb2efd [moe] implement transit between non moe tp and ep botbw 2024-07-08 09:59:46 +00:00
  • 37443cc7e4 [test] pass mixtral shardformer test botbw 2024-07-08 05:13:49 +00:00
  • 46c069b0db [zero] solve hang hxwang 2024-07-05 07:19:37 +00:00
  • 0fad23c691 [chore] handle non member group hxwang 2024-07-05 07:03:45 +00:00
  • a249e71946 [test] mixtra pp shard test hxwang 2024-07-04 06:39:01 +00:00
  • 8ae8525bdf [moe] fix plugin hxwang 2024-07-02 09:09:00 +00:00
  • 0b76b57cd6 [test] add mixtral transformer test hxwang 2024-07-02 09:08:41 +00:00
  • f9b6fcf81f [test] add mixtral for sequence classification hxwang 2024-07-02 09:02:21 +00:00
  • 1aeb5e8847 [hotfix] Remove unused plan section (#5957) Tong Li 2024-07-31 17:47:46 +08:00
  • 66fbf2ecb7 Update README.md (#5958) YeAnbang 2024-07-31 17:44:09 +08:00
  • 30f4e31a33 [Chat] Fix lora (#5946) YeAnbang 2024-07-31 14:10:17 +08:00
  • 09c5f72595 [release] update version (#5952) v0.4.2 Hongxin Liu 2024-07-31 10:04:50 +08:00
  • 060892162a [zero] hotfix update master params (#5951) Hongxin Liu 2024-07-30 13:36:00 +08:00
  • bcf0181ecd [Feat] Distrifusion Acceleration Support for Diffusion Inference (#5895) Runyu Lu 2024-07-30 10:43:26 +08:00
  • 7b38964e3a [shardformer] hotfix attn mask (#5947) Hongxin Liu 2024-07-29 19:10:06 +08:00