Commits

Breadcrumbs

History for

onmain

User selector

All users

All time

Commit history

Commits on Feb 23, 2025

• 🐈 Bye bye chat (#2934)

  Show description for f69707d
  

  qgallouedec

  and

  github-actions[bot]

  authored
  f69707dView commit details

  Copy full SHA for f69707d

  View code at this pointBrowse repository at this point

• 🧼 Upgrade ruff (#2938)

  qgallouedec

  authored
  8453017View commit details

  Copy full SHA for 8453017

  View code at this pointBrowse repository at this point

Commits on Feb 21, 2025

• 🔹 Fix: Miscalculated mask shape in comments (#2925)

  linkedlist771

  authored
  013d360View commit details

  Copy full SHA for 013d360

  View code at this pointBrowse repository at this point

Commits on Feb 20, 2025

• 🐦‍🔥 6x faster GRPO with multi-step optimization (#2899)

  Show description for e5ae703
  qgallouedec

  authored
  e5ae703View commit details

  Copy full SHA for e5ae703

  View code at this pointBrowse repository at this point

• 🪪 Adds profiling decorators for GRPOTrainer (#2889)

  Show description for a92e00e
  

  edbeeching

  and

  qgallouedec

  authored
  a92e00eView commit details

  Copy full SHA for a92e00e

  View code at this pointBrowse repository at this point

Commits on Feb 18, 2025

• 📍 [GRPO] add gradient_checkpointing (#2848)

  Show description for 9b3c5bf
  

  kashif

  and

  qgallouedec

  authored
  9b3c5bfView commit details

  Copy full SHA for 9b3c5bf

  View code at this pointBrowse repository at this point

• 🍃 GRPO - Do not load reference model when beta == 0 (#2806)

  Show description for 15fec31
  

  ingambe

  and

  qgallouedec

  authored
  15fec31View commit details

  Copy full SHA for 15fec31

  View code at this pointBrowse repository at this point

• 🩳 `max_seq_length` to `max_length` (#2895)

  Show description for be1e340
  qgallouedec

  authored
  be1e340View commit details

  Copy full SHA for be1e340

  View code at this pointBrowse repository at this point

• ⚰️ Remove deprecated (#2894)

  qgallouedec

  authored
  6aaf379View commit details

  Copy full SHA for 6aaf379

  View code at this pointBrowse repository at this point

• ✨ Add vLLM guided decoding support to GRPO Trainer (#2811)

  Show description for 49adf74
  

  kldzj

  and

  qgallouedec

  authored
  49adf74View commit details

  Copy full SHA for 49adf74

  View code at this pointBrowse repository at this point

• 🪂 Don't gather logits in SFT to avoid hanging (#2890)

  Show description for 6c54f02
  qgallouedec

  authored
  6c54f02View commit details

  Copy full SHA for 6c54f02

  View code at this pointBrowse repository at this point

• Optimize vllm num_generations (#2855)

  Show description for 963243a
  edbeeching

  authored
  963243aView commit details

  Copy full SHA for 963243a

  View code at this pointBrowse repository at this point

• 🍟 [SFT] Handles the dataset if it has been preprocessed (#2863)

  Show description for aafd8cb
  

  3 peopleauthored
  aafd8cbView commit details

  Copy full SHA for aafd8cb

  View code at this pointBrowse repository at this point

Commits on Feb 17, 2025

• 🧶 [GRPO][vLLM + LoRA] Move unmerge of PEFT model after weight loading (#2873)

  XZ-X

  authored
  8226538View commit details

  Copy full SHA for 8226538

  View code at this pointBrowse repository at this point

• 💬 Add `maybe_convert_to_chatml` map for conversational datasets in SFT (#2862)

  Show description for ba03657
  

  kashif

  and

  qgallouedec

  authored
  ba03657View commit details

  Copy full SHA for ba03657

  View code at this pointBrowse repository at this point

• [GRPO] Fix loss normalization (#2881)

  Show description for 293b620
  edbeeching

  authored
  293b620View commit details

  Copy full SHA for 293b620

  View code at this pointBrowse repository at this point

• [SFT] fix check for AutoLigerKernelForCausalLM (#2874)

  Show description for 6d9fc11
  kashif

  authored
  6d9fc11View commit details

  Copy full SHA for 6d9fc11

  View code at this pointBrowse repository at this point

Commits on Feb 13, 2025

• ⬆️ Bump dev version

  qgallouedec

  committed
  ffcb9f4View commit details

  Copy full SHA for ffcb9f4

  View code at this pointBrowse repository at this point

• Release: v0.15

  qgallouedec

  committed
  00e5889View commit details

  Copy full SHA for 00e5889

  View code at this pointBrowse repository at this point

• 👨‍👩‍👧 GRPO + PEFT + vLLM (#2818)

  Show description for 5c9cf20
  

  winglian

  and

  qgallouedec

  authored
  5c9cf20View commit details

  Copy full SHA for 5c9cf20

  View code at this pointBrowse repository at this point

• 🪆 Fix for Incorrect ValueError Handling in reward_weights in grpo_trainer.py (#2843)

  Show description for 8830786
  loveychen

  authored
  8830786View commit details

  Copy full SHA for 8830786

  View code at this pointBrowse repository at this point

• Fix PeftModel check when moving weights to vlllm (#2850)

  Show description for b0f513c
  edbeeching

  authored
  b0f513cView commit details

  Copy full SHA for b0f513c

  View code at this pointBrowse repository at this point

Commits on Feb 12, 2025

• ⚡ Fix GRPO PEFT (#2725)

  qgallouedec

  authored
  8122166View commit details

  Copy full SHA for 8122166

  View code at this pointBrowse repository at this point

Commits on Feb 11, 2025

• 🥾 Allow bootstrap GRPO (#2829)

  Show description for 7347c29
  

  qgallouedec

  and

  kashif

  authored
  7347c29View commit details

  Copy full SHA for 7347c29

  View code at this pointBrowse repository at this point

• 🙌 Share vLLM device with training when only 1 available (#2827)

  Show description for 9b67eea
  qgallouedec

  authored
  9b67eeaView commit details

  Copy full SHA for 9b67eea

  View code at this pointBrowse repository at this point

• ⚖️ Add reward weight in multi-reward settings for GRPO (#2676)

  Show description for e752fc6
  

  hesamsheikh

  and

  qgallouedec

  authored
  e752fc6View commit details

  Copy full SHA for e752fc6

  View code at this pointBrowse repository at this point

• 🫘 Add `set_seed()` call in GRPO to ensure unique seed for each process (#2824)

  Show description for 674bb75
  qgallouedec

  authored
  674bb75View commit details

  Copy full SHA for 674bb75

  View code at this pointBrowse repository at this point

Commits on Feb 10, 2025

• 📤 GRPO refactor loading the model weights to vllm (#2817)

  Show description for b9df810
  

  winglian

  and

  qgallouedec

  authored
  b9df810View commit details

  Copy full SHA for b9df810

  View code at this pointBrowse repository at this point

Commits on Feb 7, 2025

• ⛰️ Reduce peak vram consumption with efficient selective log_softmax (#2799)

  Show description for 09eefa7
  

  tyler-romero

  and

  qgallouedec

  authored
  09eefa7View commit details

  Copy full SHA for 09eefa7

  View code at this pointBrowse repository at this point

• 🔬 SFT simplification (#2405)

  Show description for 5b9236d
  

  qgallouedec

  and

  kashif

  authored
  5b9236dView commit details

  Copy full SHA for 5b9236d

  View code at this pointBrowse repository at this point

• 📠 Log completions for GRPO (#2772)

  Show description for 82d12eb
  

  qgallouedec

  and

  lewtun

  authored
  82d12ebView commit details

  Copy full SHA for 82d12eb

  View code at this pointBrowse repository at this point

• 🎯 [SFT] add token accuracy metric (#2597)

  Show description for 84d73fd
  

  kashif

  and

  qgallouedec

  authored
  84d73fdView commit details

  Copy full SHA for 84d73fd

  View code at this pointBrowse repository at this point

• 🆚 Distinguish padding and eos when they differ (#2793)

  binary-husky

  authored
  2241f17View commit details

  Copy full SHA for 2241f17

  View code at this pointBrowse repository at this point

Commits on Feb 6, 2025

• 📉 Optimize GRPO memory usage by redefining `per_device_batch_size` as generations per device (#2776)

  Show description for cf97133
  qgallouedec

  authored
  cf97133View commit details

  Copy full SHA for cf97133

  View code at this pointBrowse repository at this point

• Revert "Before the first training step, the model has no optimizer: fix ds3"

  Show description for 7134a1e
  qgallouedec

  committed
  7134a1eView commit details

  Copy full SHA for 7134a1e

  View code at this pointBrowse repository at this point