Commits

Branch selector

main

User selector

All users

All time

Commit history

Commits on Mar 5, 2025

• 🚀 Supporting `deepspeed>=0.16.4`'s rename (#2963)

  Show description for e3244d2
  jamesbraza

  authored
  e3244d2View commit details

  Copy full SHA for e3244d2

  Browse repository at this point

• 🎲 Add support for additional generation kwargs in GRPO Trainer (#2989)

  Show description for 6a02c69
  

  3 peopleauthored
  6a02c69View commit details

  Copy full SHA for 6a02c69

  Browse repository at this point

Commits on Mar 4, 2025

• 🗜️ Loosened tokenizer type hint on `apply_chat_template` (#3005)

  jamesbraza

  authored
  a1c58aaView commit details

  Copy full SHA for a1c58aa

  Browse repository at this point

• 🌍 Use global normalization for KL logging (to match normalization for loss) (#3004)

  Show description for 3f0695a
  

  tchang1997

  and

  qgallouedec

  authored
  3f0695aView commit details

  Copy full SHA for 3f0695a

  Browse repository at this point

• 📚 Update customization and distributing training documentation (#2991)

  qgallouedec

  authored
  a72b50bView commit details

  Copy full SHA for a72b50b

  Browse repository at this point

• ✌️ Remove double compute of sum in SFTTrainer (#3001)

  Show description for ea1d9be
  

  lexasub

  and

  qgallouedec

  authored
  ea1d9beView commit details

  Copy full SHA for ea1d9be

  Browse repository at this point

• Improve ci (#3007)

  Show description for 402187b
  paulinebm

  authored
  402187bView commit details

  Copy full SHA for 402187b

  Browse repository at this point

• 🪙 [SFT] Log `num_tokens` and some logging fixes (#3006)

  qgallouedec

  authored
  5858ceaView commit details

  Copy full SHA for 5858cea

  Browse repository at this point

Commits on Mar 3, 2025

• Update pr_style_bot.yml (#3003)

  qgallouedec

  authored
  7442d42View commit details

  Copy full SHA for 7442d42

  Browse repository at this point

• 🚀 DeepSpeed integration documentation (#2993)

  Show description for 98de0e7
  

  qgallouedec

  and

  kashif

  authored
  98de0e7View commit details

  Copy full SHA for 98de0e7

  Browse repository at this point

Commits on Feb 28, 2025

• 🛣️ `inference_mode` to `no_grad` when computing `old_per_token_logps` (#2987)

  qgallouedec

  authored
  491921cView commit details

  Copy full SHA for 491921c

  Browse repository at this point

• 🫔 [GRPO] Pass wrapped model to unwrap_model_for_generation for DeepSpeed Stage-3 compatibility (#2871)

  Show description for ad6a35b
  

  kiddj

  and

  qgallouedec

  authored
  ad6a35bView commit details

  Copy full SHA for ad6a35b

  Browse repository at this point

• 🔍 Update GRPO config documentation for beta parameter stability (#2992)

  nopepper

  authored
  7bc9858View commit details

  Copy full SHA for 7bc9858

  Browse repository at this point

• ⚰️ Deprecate liger-kernel (#2949)

  Show description for b882f57
  

  qgallouedec

  and

  kashif

  authored
  b882f57View commit details

  Copy full SHA for b882f57

  Browse repository at this point

• 📑 Fix logged metrics for KTO (#2982)

  vaibhavjindal

  authored
  ac7bde5View commit details

  Copy full SHA for ac7bde5

  Browse repository at this point

• 📜 Update README and doc index (#2986)

  Show description for 3d94e4e
  qgallouedec

  authored
  3d94e4eView commit details

  Copy full SHA for 3d94e4e

  Browse repository at this point

• 🧬 Fix typo in grpo_trainer.py (#2988)

  congchan

  authored
  1a303ccView commit details

  Copy full SHA for 1a303cc

  Browse repository at this point

Commits on Feb 27, 2025

• 🪪 Adds a more fine-grained profiling context (#2975)

  Show description for ac327d5
  

  edbeeching

  and

  qgallouedec

  authored
  ac327d5View commit details

  Copy full SHA for ac327d5

  Browse repository at this point

• 🌌 Fix logits computation in trainer prediction step (#2969)

  Show description for c0854c3
  

  4 peopleauthored
  c0854c3View commit details

  Copy full SHA for c0854c3

  Browse repository at this point

• 👂 Update learning rate doc in `KTOConfig` (#2912)

  Show description for aa18ecf
  

  sileod

  and

  qgallouedec

  authored
  aa18ecfView commit details

  Copy full SHA for aa18ecf

  Browse repository at this point

• 🕸 Add distributing training guide (#2956)

  qgallouedec

  authored
  6849c05View commit details

  Copy full SHA for 6849c05

  Browse repository at this point

• 🧗 Add GRPO Trainer support for third-party accelerators (#2836)

  Show description for 27a6f22
  

  ji-huazhong

  and

  qgallouedec

  authored
  27a6f22View commit details

  Copy full SHA for 27a6f22

  Browse repository at this point

• 👧🏽 Adding DoRA support to model config (#2974)

  nbasyl

  authored
  f074dcdView commit details

  Copy full SHA for f074dcd

  Browse repository at this point

• Update grpo_trainer.py (#2973)

  tpoisonooo

  authored
  0caff61View commit details

  Copy full SHA for 0caff61

  Browse repository at this point

Commits on Feb 26, 2025

• 🔢 Fix GRPO doc about `num_iterations` (#2966)

  qgallouedec

  authored
  019fc6dView commit details

  Copy full SHA for 019fc6d

  Browse repository at this point

Commits on Feb 24, 2025

• ⏪ Parameterize `enable_prefix_caching` (#2900)

  Show description for 69ad852
  

  ji-huazhong

  and

  qgallouedec

  authored
  69ad852View commit details

  Copy full SHA for 69ad852

  Browse repository at this point

• 📌 Pin liger-kernel and vLLM (#2952)

  Show description for 45ccdef
  qgallouedec

  authored
  45ccdefView commit details

  Copy full SHA for 45ccdef

  Browse repository at this point

• 🗿 Updated DPO default values for alpha and tau (#2918)

  Show description for 703484a
  

  Ishan-Kumar2

  and

  qgallouedec

  authored
  703484aView commit details

  Copy full SHA for 703484a

  Browse repository at this point

• ↩️ Fix typo in TextEnvironment init param, should be max_tool_response (#2921)

  Show description for 9b76d5f
  

  shenxiangzhuang

  and

  qgallouedec

  authored
  9b76d5fView commit details

  Copy full SHA for 9b76d5f

  Browse repository at this point

• 📇 GRPO: print completions to console and update docs (#2951)

  Show description for cbe0681
  

  3 peopleauthored
  cbe0681View commit details

  Copy full SHA for cbe0681

  Browse repository at this point

• ✋ Prevent applying the chat template to tokenized datasets (#2939)

  Show description for 4e0cf01
  

  3 peopleauthored
  4e0cf01View commit details

  Copy full SHA for 4e0cf01

  Browse repository at this point

• 🐯 Fix LigerKernel for SFTTrainer (#2940)

  Show description for 5c05913
  

  lewtun

  and

  qgallouedec

  authored
  5c05913View commit details

  Copy full SHA for 5c05913

  Browse repository at this point

• ☠️ Update `max_seq_length` to `max_length` in `SFTConfig` (#2947)

  qgallouedec

  authored
  caba04dView commit details

  Copy full SHA for caba04d

  Browse repository at this point

• 📋 Add vLLM version to environment printout (#2946)

  qgallouedec

  authored
  be5a088View commit details

  Copy full SHA for be5a088

  Browse repository at this point

• ♻️ Fix caching in SFT (#2945)

  qgallouedec

  authored
  3886147View commit details

  Copy full SHA for 3886147

  Browse repository at this point