trending/miles
1
0
Fork 0
mirror of https://github.com/radixark/miles.git synced 2026-10-02 07:14:53 +08:00
Code Issues Packages Projects Releases Wiki Activity
2,631 Commits 2,591 Branches 50 Tags
main
Commit Graph
100 Commits
This Branch
This Branch
All Branches
Author SHA1 Message Date
Zilin Zhu a961dfddde [docker] update dockerfile and some code cleanup 2025-10-05 16:14:02 +00:00
Zilin Zhu de080f45df [xtuner] remove xtuner backend (#424) 
As we have an almost finished FSDP backend and a hybrid hf + megatron backend, it seems unnecessary to have xtuner backend
 2025-10-05 17:35:04 +08:00
Zilin Zhu c181fe23e5 [doc] update docs 2025-10-05 09:05:57 +00:00
Zilin Zhu 0beb909eb4 [doc] update docs 2025-10-05 08:55:27 +00:00
Zilin Zhu 5336d85759 [doc] update docs 2025-10-05 08:49:50 +00:00
Zilin Zhu c143ca2b23 [docker] patch offload bugfix on Qwen3Next in sglang 2025-10-04 14:44:06 +00:00
Zilin Zhu ab13a4aed8 [model] Support Qwen3-Next-80B-A3B (#417) 
* [model] Support Qwen3-Next-80B-A3B

* bugfix

* use fla's ShortConvolution and support data packing
 2025-10-04 22:12:07 +08:00
Zilin Zhu f291a8f5a5 add ad-hoc wait for large moe 2025-09-30 11:07:56 +00:00
Zilin Zhu dde524e2c6 [feat] support fault tolerant for rollout engines (#405) 
* [feat] support fault tolerant for rollout engines

* support fault tolerant for UpdateWeightFromDistributed

* bugfix

* bugfix
 2025-09-30 14:18:13 +08:00
Zilin Zhu 584657a585 [refactor] remove Registry and change the order of init (#398) 
* [refactor] remove Registry and change the order of init

* bugfix
 2025-09-28 22:20:52 +08:00
Zilin Zhu eb5604e20f [docker] fix routing replay with pp and other bugfixes (#395) 2025-09-28 14:55:24 +08:00
Zilin Zhu d862870c6a remove slime/tests folder 2025-09-27 01:34:53 +00:00
Zilin Zhu 5b77a9b2f5 [feat] add --use-routing-replay (#387) 2025-09-26 20:38:56 +08:00
Zilin Zhu 921e2656aa a strange patch to prevent the sglang server not immediately offloaded 2025-09-24 04:18:36 -07:00
Zilin Zhu 92a8b79b77 [feat] enable use_flattened_tensor_bucket with quantization config (#374) 2025-09-24 16:10:08 +08:00
Zilin Zhu eb0aa92012 update docs 2025-09-23 09:34:15 -07:00
Zilin Zhu 4be82aa9cb [doc] update doc 2025-09-23 07:11:39 -07:00
Zilin Zhu 69734956e1 [reproducibility][docker] enable training reproducibility (#370) 
* [reproducibility][docker] enable training reproducibility

* bugfix

* bugfix

* add example

* update docs
 2025-09-23 20:20:02 +08:00
Zilin Zhu ddfadcc46d add fall through to sglang_enable_deterministic_inference 2025-09-22 22:31:05 -07:00
Zilin Zhu 52e5920432 [bugfix] fix eval_reward_key 2025-09-22 20:23:39 -07:00
Zilin Zhu 4264d2c365 [router] add back /retrieve_from_text 2025-09-22 18:09:51 -07:00
Zilin Zhu 212f0d333e [feat] support distributed post to enable more concurrent requests (#368) 2025-09-23 08:03:59 +08:00
Zilin Zhu 6faae2f5dc [router] extract middleware folder (#367) 2025-09-22 22:24:36 +08:00
Zilin Zhu dd2503bfb7 [router] support slime-router only (#366) 2025-09-22 21:04:29 +08:00
Zilin Zhu ba0b13a5e7 [bugfix] fix the num_tokens used for per_token_loss in multi-turn training (#365) 
* [doc] add bibtext

* [bugfix] fix the num_tokens used for per_token_loss in multi-turn training
 2025-09-22 15:31:09 +08:00
Zilin Zhu 815e8392f9 [doc] add bibtext 2025-09-21 20:18:45 -07:00
Zilin Zhu 758ed2494e [refactor] Add actor registry (#359) 2025-09-20 00:23:07 +08:00
Zilin Zhu 843c94e9a0 [feat] add --critic-lr and --num-critic-only-steps (#350) 2025-09-19 23:25:48 +08:00
Zilin Zhu 6e039591cb [docker] move zhuzilin/slime to slimerl/slime 2025-09-17 22:28:46 -07:00
Zilin Zhu 17f8762996 fix lint 2025-09-17 02:55:03 -07:00
Zilin Zhu 4badb749b2 [feat] init support for PPO (#342) 2025-09-17 08:13:22 +08:00
Zilin Zhu 1785a8d43d [nfc] remove with_optimizer as it's not used 2025-09-15 04:27:19 -07:00
Zilin Zhu 2a3a2b6752 [feat] use one global httpx.AsyncClient and remove --use-http2 (#338) 
* [feat] use one global httpx.AsyncClient

* remove --use-http2
 2025-09-15 19:01:47 +08:00
Zilin Zhu aa3375040b [model] fix qwen3-14B.sh 2025-09-12 02:50:19 -07:00
Zilin Zhu 1ff6398e55 minor fix for config mapping 2025-09-11 05:15:13 -07:00
Zilin Zhu 32d55a634f [refactor] remove slime/backend/utils and extract slime_validate_args (#329) 2025-09-11 19:46:20 +08:00
Zilin Zhu 5952fa3864 [bugfix] fix no weight_version for aborted samples (#327) 2025-09-11 10:47:57 +08:00
Zilin Zhu d22225f622 [xtuner] add wandb (#324) 2025-09-11 00:23:05 +08:00
Zilin Zhu b696394e6a [xtuner] unroll TrainingWorker and TrainEngine (#322) 
* [xtuner] unroll TrainingWorker and TrainEngine

* unroll _packing

* remove unnecessary func

* only print on rank0
 2025-09-10 20:26:40 +08:00
Zilin Zhu ebf16c57c2 [refactor] bind numa and rename num_gpus_per_node (#316) 2025-09-09 19:58:50 +08:00
Zilin Zhu 3f5ddd8c0c [nfc] cleanup for weight_version (#314) 2025-09-09 16:04:46 +08:00
Zilin Zhu 1099805fe3 [docker] update to sglang 0.5.2rc2 (#313) 2025-09-09 15:19:41 +08:00
Zilin Zhu 17ac8076d9 [feat] init xtuner backend (#310) 2025-09-09 01:00:05 +08:00
Zilin Zhu 982116eaff [scripts] update run-qwen3-235B-A22B.sh for weight updation benchmark 2025-09-08 02:57:44 -07:00
Zilin Zhu d4a774126f [doc] fix typos 2025-09-06 10:43:02 -07:00
Zilin Zhu 741f24a995 Revert "Website entry for Sphinx style docs (#284)" (#286) 
This reverts commit f90bb2793c.
 2025-09-06 11:15:01 +08:00
Zilin Zhu cfad46fc22 [feat] init support for FSDP (#282) 
* [feat] init support for FSDP

* bugfix

* bugfix
 2025-09-06 01:03:18 +08:00
Zilin Zhu c64c26062a [refactor] remove over_sampling_filter and extract some functions (#278) 2025-09-05 13:59:42 +08:00
Zilin Zhu 675ca6e75a [refactor] Add isort back and move global gloo to global util (#273) 2025-09-05 11:40:48 +08:00
Zilin Zhu cd3da9db64 [doc] bugfix 2025-09-04 03:42:57 -07:00
Zilin Zhu 00b01c7584 fix typo 2025-09-04 03:19:15 -07:00
Zilin Zhu 547953aec0 [doc] add mult_agent example to doc 2025-09-04 03:14:01 -07:00
Zilin Zhu e33a85bae1 [doc] prettifiy language convertion toggle (#270) 
* [doc] prettifiy language convertion toggle

* fix lint
 2025-09-04 18:09:54 +08:00
Zilin Zhu 5de8a67f24 [doc] add blogs (#268) 2025-09-04 17:29:30 +08:00
Zilin Zhu 8255c9b62e [doc] add example doc to the website (#267) 2025-09-04 16:27:46 +08:00
Zilin Zhu d2e14ad913 [doc] reorg doc 2025-09-03 23:50:54 -07:00
Zilin Zhu 4c2c763b8c [doc] rearrange doc 2025-09-03 23:34:18 -07:00
Zilin Zhu 62d104c55e [doc] fix link 2025-09-03 23:11:29 -07:00
Zilin Zhu a9d59b71d9 [bugfix] fix multi agent bugs (#260) 2025-09-04 13:52:33 +08:00
Zilin Zhu 646d9808ad [doc] add examples folder link to readme 2025-09-03 20:01:09 -07:00
Zilin Zhu 8f4bbafca3 [example] Add fully async example (#258) 
* [example] Add fully async example

* add README
 2025-09-04 10:57:50 +08:00
Zilin Zhu d49e45f032 Revert "[bugfix] fix read data containing "tools" field (#255)" (#256) 
This reverts commit 5c4fce60fc.
 2025-09-03 22:36:59 +08:00
Zilin Zhu 7c0a0341d5 [bugfix] fix weight convertion for glm4 2025-09-03 05:30:31 -07:00
Zilin Zhu 4fd4f2735d [bugfix] prevent empty param bucket 2025-09-03 02:29:09 -07:00
Zilin Zhu 81c1f8661a [docker] bugfix for sglang dp attention 2025-09-02 22:59:04 -07:00
Zilin Zhu 9a291fca71 [bugfix] use fp32 for rollout_log_probs (#245) 2025-09-02 17:54:39 +08:00
Zilin Zhu 261ecee700 [release] bump to v0.1.0 (#239) 
* [release] bump to v0.1.0

* only run actions on the latest commit

* run actions only on pr

* fix lint

* remove mamba self-update

* fix lint

* bugfix

* fix sglang commit

* bugfix

* bugfix
 2025-09-01 00:28:28 +08:00
Zilin Zhu 1e68dc8c19 [nfc] code cleanup (#240) 
* [nfc] code cleanup

* rename ppo_actor.py to train_actor.py
 2025-08-30 20:06:58 +08:00
Zilin Zhu 0b96b53cc4 [tests] remove and rename tests 2025-08-29 14:02:46 -07:00
Zilin Zhu 0e1c4b2e2f [doc] some clean up for the doc (#238) 
* [doc] some clean up for the doc

* remove redundant scripts
 2025-08-30 04:57:05 +08:00
Zilin Zhu fd7c64e846 [feature] add --tis-clip-low (#236) 2025-08-29 17:24:23 +08:00
Zilin Zhu 868a3db256 set ref_ckpt_step to ckpt_step when load is not set 2025-08-29 01:10:08 -07:00
Zilin Zhu e35f515758 [docker] set 0.5.0rc0 as new stable release 2025-08-29 00:22:25 -07:00
Zilin Zhu cbe57433fd skip megatron check and init when debug_rollout_only 2025-08-28 22:46:40 -07:00
Zilin Zhu 0eff732497 bugfix on deepep init without offloading 2025-08-28 19:32:01 -07:00
Zilin Zhu 552ba47f5d use rollout/step for perf 2025-08-28 18:02:45 -07:00
Zilin Zhu fd31dc7563 [profile] add perf stats on rollout 2025-08-28 04:48:22 -07:00
Zilin Zhu 0070e590fe [feat] add --ref-ckpt-step 2025-08-28 03:31:58 -07:00
Zilin Zhu e74666b674 [refactor] refactor tis implementation (#234) 
* [refactor] refactor tis implementation

* bugfix
 2025-08-28 16:17:01 +08:00
Zilin Zhu b738d3338a [ci] init gpu ci (#231) 2025-08-27 12:45:30 +08:00
Zilin Zhu aea44777ab [test] add e2e test for quick start 2025-08-26 01:25:04 -07:00
Zilin Zhu 771e0f2504 [docker] fix megatron patch 2025-08-26 00:05:53 -07:00
Zilin Zhu 529e6e2c0f [docker] update to latest megatron 2025-08-25 20:11:45 -07:00
Zilin Zhu 81fdfa2a7a [docker] update patch for megatron deepep 2025-08-25 00:45:10 -07:00
Zilin Zhu b1e2c65f68 [feat] support generate multi samples for one prompt (#216) 2025-08-25 12:12:55 +08:00
Zilin Zhu e943681211 [feat] support training with sglang draft model without mtp (#204) 
* [feat] support training with sglang draft model without mtp

* fix lint

* fix lint
 2025-08-20 21:07:42 +08:00
Zilin Zhu d3306394d2 bugfix for gspo with long prompt 2025-08-20 03:25:36 -07:00
Zilin Zhu 8e994cecba always cpu actor weights after training 2025-08-17 19:57:43 -07:00
Zilin Zhu 968f6deb10 use token-in token-out as default (#191) 2025-08-15 11:47:15 +08:00
Zilin Zhu 2275717282 [script] increase --sglang-mem-fraction-static to 0.7 2025-08-14 20:28:57 -07:00
Zilin Zhu 6ae99c3352 Revert "Minor extract generate one sample (#168)" (#190) 
This reverts commit fe5e4e95bf.
 2025-08-14 19:23:01 +08:00
Zilin Zhu 9da9d55e81 bugfix 2025-08-14 01:53:10 -07:00
Zilin Zhu c36a0142f7 fix README 2025-08-14 01:11:02 -07:00
Zilin Zhu 3b4eeabf3f [refactor] rename buffer to rollout controller and extract buffer to data source (#189) 2025-08-14 16:06:26 +08:00
Zilin Zhu dfd91a58ef Revert "Add new API for rollout_fn (#169)" (#188) 
This reverts commit 4c253edb9d.
 2025-08-14 15:08:16 +08:00
Zilin Zhu 70564d9f5c Revert "Extract buffer logic to legacy adapter rollout fn (#171)" (#187) 
This reverts commit 118b6a02b4.
 2025-08-14 14:56:53 +08:00
Zilin Zhu d0712a50bb Revert "Implement PartialRolloutFn abstraction for clean partial rollout (#172)" (#186) 
This reverts commit 2a95a217c2.
 2025-08-14 14:56:24 +08:00
Zilin Zhu a96a02e9d1 Revert "Remove singleton GenerateState (#173)" (#185) 
This reverts commit 27ef0076f3.
 2025-08-14 14:55:39 +08:00
Zilin Zhu fdb203b917 Revert "Remove several fields from global rollout state (#174)" (#184) 
This reverts commit 51dd193d0b.
 2025-08-14 14:54:33 +08:00
Zilin Zhu 0197de44f4 Revert "Minor cleanup sglang_rollout (#175)" (#183) 
This reverts commit 2dbeb3a44c.
 2025-08-14 14:53:35 +08:00