Commits

Breadcrumbs

History for

onmain

User selector

All users

All time

Commit history

Commits on Feb 5, 2025

• [misc] fix: load and offload in compute log prob (#208)

  Show description for 6872dbe
  PeterSH6

  authored
  6872dbeView commit details

  Copy full SHA for 6872dbe

  View code at this pointBrowse repository at this point

• fix critic save error (#199)

  Show description for 4669ab9
  WeiXiongUST

  authored
  4669ab9View commit details

  Copy full SHA for 4669ab9

  View code at this pointBrowse repository at this point

Commits on Feb 3, 2025

• megatron: fix config error and add compute log prob interface (#186)

  eric-haibin-lin

  authored
  818e4deView commit details

  Copy full SHA for 818e4de

  View code at this pointBrowse repository at this point

Commits on Feb 2, 2025

• exchange the mini_batch_size calculation logic (#183)

  chujiezheng

  authored
  6e003c0View commit details

  Copy full SHA for 6e003c0

  View code at this pointBrowse repository at this point

Commits on Feb 1, 2025

• [misc] fix: grpo kl loss should be add when do minimization (#179)

  Show description for a65c915
  PeterSH6

  authored
  a65c915View commit details

  Copy full SHA for a65c915

  View code at this pointBrowse repository at this point

Commits on Jan 31, 2025

• fix bug：fix checkpoint save with existing dirs (#174)

  Show description for 679798c
  

  dignfei

  and

  eric-haibin-lin

  authored
  679798cView commit details

  Copy full SHA for 679798c

  View code at this pointBrowse repository at this point

Commits on Jan 30, 2025

• [Liger-kernel] Add an option to use `_apply_liger_kernel_to_instance()` to load model (#133)

  Show description for dd41877
  hongpeng-guo

  authored
  dd41877View commit details

  Copy full SHA for dd41877

  View code at this pointBrowse repository at this point

• [nit] Explicitly show logits dimension in `dp_worker.py::forward_micro_batch` (#164)

  Show description for df03aa6
  hongpeng-guo

  authored
  df03aa6View commit details

  Copy full SHA for df03aa6

  View code at this pointBrowse repository at this point

Commits on Jan 27, 2025

• [misc] fix: gradient accumulation in seq balance and modify default vllm log level (#141)

  Show description for 695bdbb
  PeterSH6

  authored
  695bdbbView commit details

  Copy full SHA for 695bdbb

  View code at this pointBrowse repository at this point

• [VLLM] Set max_num_batched_tokens for vllm rollout (#140)

  Show description for c99df03
  xingyaoww

  authored
  c99df03View commit details

  Copy full SHA for c99df03

  View code at this pointBrowse repository at this point

• [BREAKING][misc] feat: change micro_batch_size to micro_batch_size_per_gpu (#136)

  Show description for f2a76ac
  PeterSH6

  authored
  f2a76acView commit details

  Copy full SHA for f2a76ac

  View code at this pointBrowse repository at this point

• [misc] fix: only return old_log_prob and temp to fix union problem in old_log_prob (#137)

  Show description for 8bf9b95
  PeterSH6

  authored
  8bf9b95View commit details

  Copy full SHA for 8bf9b95

  View code at this pointBrowse repository at this point

Commits on Jan 24, 2025

• [ppo] refactor: refactor old_log_prob into a separate function (#129)

  vermouth1992

  authored
  e974fb1View commit details

  Copy full SHA for e974fb1

  View code at this pointBrowse repository at this point

• [perf] feat: support ref/rm offload (#121)

  Show description for fefca41
  vermouth1992

  authored
  fefca41View commit details

  Copy full SHA for fefca41

  View code at this pointBrowse repository at this point

Commits on Jan 23, 2025

• [algo] feat: support GRPO algorithm (#124)

  Show description for cd52d8b
  PeterSH6

  authored
  cd52d8bView commit details

  Copy full SHA for cd52d8b

  View code at this pointBrowse repository at this point

• [misc] fix: normalize batch size should divide sp size (#125)

  Show description for 5b90cd7
  PeterSH6

  authored
  5b90cd7View commit details

  Copy full SHA for 5b90cd7

  View code at this pointBrowse repository at this point

Commits on Jan 21, 2025

• [perf] feat: Support dynamic batch size (#118)

  Show description for 54051c8
  

  vermouth1992

  and

  PeterSH6

  authored
  54051c8View commit details

  Copy full SHA for 54051c8

  View code at this pointBrowse repository at this point

Commits on Jan 19, 2025

• [misc] feat: support mfu calculation (#117)

  vermouth1992

  authored
  41f645dView commit details

  Copy full SHA for 41f645d

  View code at this pointBrowse repository at this point

Commits on Jan 18, 2025

• [perf] fix: set use_reentrant=False when enable gradient checkpointing (#114)

  Show description for 5a94e14
  vermouth1992

  authored
  5a94e14View commit details

  Copy full SHA for 5a94e14

  View code at this pointBrowse repository at this point

• [misc][Long Context] feat: support ulysses for long context training (#109)

  PeterSH6

  authored
  e8eb9e4View commit details

  Copy full SHA for e8eb9e4

  View code at this pointBrowse repository at this point

Commits on Jan 14, 2025

• refact: hybrid_engine dir to sharding_manager for more general representation (#103)

  PeterSH6

  authored
  6a9f6e1View commit details

  Copy full SHA for 6a9f6e1

  View code at this pointBrowse repository at this point

Commits on Jan 13, 2025

• [misc] feat: support different flash_attn versions with variable num returns (#100)

  Show description for 1facb9d
  PeterSH6

  authored
  1facb9dView commit details

  Copy full SHA for 1facb9d

  View code at this pointBrowse repository at this point

• [misc] fix reward model issue with TokenClassification model and support running particular steps instead of epochs (#99)

  Show description for a0e8ed2
  PeterSH6

  authored
  a0e8ed2View commit details

  Copy full SHA for a0e8ed2

  View code at this pointBrowse repository at this point

Commits on Jan 11, 2025

• [misc] feat: spport rmpad/data-packing in FSDP with transformers (#91)

  Show description for 569210e
  PeterSH6

  authored
  569210eView commit details

  Copy full SHA for 569210e

  View code at this pointBrowse repository at this point

Commits on Jan 5, 2025

• [rollout] feat: support best-of-n generation in vLLM (#80)

  PeterSH6

  authored
  2997e5dView commit details

  Copy full SHA for 2997e5d

  View code at this pointBrowse repository at this point

Commits on Jan 2, 2025

• Support Mlflow, allow forward to have different batch size, compute more metrics (#74)

  Show description for 2856be4
  fzyzcjy

  authored
  2856be4View commit details

  Copy full SHA for 2856be4

  View code at this pointBrowse repository at this point

Commits on Dec 21, 2024

• [misc] feat: remove @ray.remote on workers to allow inheritance (#61)

  Show description for 99d2c19
  

  PeterSH6

  and

  eric-haibin-lin

  authored
  99d2c19View commit details

  Copy full SHA for 99d2c19

  View code at this pointBrowse repository at this point

• [install] fix: revert pyproj.toml and fix tensordict req (#59)

  Show description for 09568e6
  eric-haibin-lin

  authored
  09568e6View commit details

  Copy full SHA for 09568e6

  View code at this pointBrowse repository at this point

Commits on Dec 18, 2024

• [ppo] chore: remove unused flash_attn dependency, and add docs for GSM8k reward (#55)

  Show description for cf01808
  eric-haibin-lin

  authored
  cf01808View commit details

  Copy full SHA for cf01808

  View code at this pointBrowse repository at this point

• [BREAKING][refact]: move actor/critic/hybrid_engine/reward_model/rollout/workers out of ppo directory for reuse (#52)

  Show description for f535977
  PeterSH6

  authored
  f535977View commit details

  Copy full SHA for f535977

  View code at this pointBrowse repository at this point

Loading