Commits

Branch selector

 main

User selector

All users

Datepicker

All time

Commit History

Commits on Sep 12, 2025

• [README] Add explanations for varlen training

  yzhangcs

  authored
  f5c618c

  Copy full SHA for f5c618c

Commits on Sep 7, 2025

• Add 1B GSA config

  yzhangcs

  authored
  0a1454e

  Copy full SHA for 0a1454e

Commits on Aug 17, 2025

• Allow mixed-precision training in DDP/single-device scenarios (#36)

  Show description for 24b2980
  mutiann

  authored
  24b2980

  Copy full SHA for 24b2980

Commits on Aug 12, 2025

• Disable fused linear CE if necessary

  yzhangcs

  authored
  6e49e20

  Copy full SHA for 6e49e20

Commits on Jun 12, 2025

• [Custom model] add document and example

  sustcsonglin

  committed
  8684660

  Copy full SHA for 8684660

Commits on May 13, 2025

• update citation

  sustcsonglin

  authored
  03d3739

  Copy full SHA for 03d3739

• Add bib for `flame`

  yzhangcs

  authored
  c43ad4d

  Copy full SHA for c43ad4d

Commits on May 7, 2025

• Delete `norm_first` option

  yzhangcs

  committed
  4790efe

  Copy full SHA for 4790efe

• Add GatedDeltaNet configs (#30)

  yzhangcs

  committed
  b654e4f

  Copy full SHA for b654e4f

• Roll back the initial std range. Use larger learning rate by default

  yzhangcs

  committed
  a587400

  Copy full SHA for a587400

• Delete all 3rdparties

  yzhangcs

  committed
  c4939ae

  Copy full SHA for c4939ae

• [Misc.] Do not use streaming for training

  yzhangcs

  committed
  28d1857

  Copy full SHA for 28d1857

Commits on Apr 27, 2025

• Save `bits_per_token` for each sample during preprocessing

  yzhangcs

  authored
  a0a1069

  Copy full SHA for a0a1069

Commits on Apr 26, 2025

• Increase default `num_workers` & add `batch_size` arg for preprocessing

  yzhangcs

  committed
  aaca529

  Copy full SHA for aaca529

• Add scripts for data preprocessing

  yzhangcs

  committed
  e5cbd0c

  Copy full SHA for e5cbd0c

• Move dataset building to `flame.data`

  yzhangcs

  committed
  2c49713

  Copy full SHA for 2c49713

Commits on Apr 25, 2025

• Add mamba configs (#22)

  yzhangcs

  committed
  d834161

  Copy full SHA for d834161

Commits on Apr 12, 2025

• Use unsigned int to save memories

  yzhangcs

  authored
  c7c0183

  Copy full SHA for c7c0183

Commits on Apr 8, 2025

• add back grad_norm metric (#25)

  Show description for fa5e448
  

  kashif

  and

  yzhangcs

  authored
  fa5e448

  Copy full SHA for fa5e448

Commits on Apr 7, 2025

• Data collator fixes (#24)

  Show description for 7f8789e
  

  kashif

  and

  yzhangcs

  authored
  7f8789e

  Copy full SHA for 7f8789e

Commits on Apr 5, 2025

• Update lint workflow to enable recursive submodule checkout

  yzhangcs

  committed
  42a7897

  Copy full SHA for 42a7897

• Fix default config path

  yzhangcs

  authored
  9bb3d9e

  Copy full SHA for 9bb3d9e

• Remove incorrect gradient_accumulation_steps scaling in trainer (#23)

  Show description for aee63f8
  kashif

  authored
  aee63f8

  Copy full SHA for aee63f8

• Update workflow actions

  yzhangcs

  committed
  0de4470

  Copy full SHA for 0de4470

• Update torchtitan and train.py (#21)

  Show description for 7f85833
  

  kashif

  and

  yzhangcs

  authored
  7f85833

  Copy full SHA for 7f85833

Commits on Mar 22, 2025

• Support saving log files

  yzhangcs

  committed
  c949efe

  Copy full SHA for c949efe

• Move taining scripts into the `flame` folder

  yzhangcs

  committed
  ebb9866

  Copy full SHA for ebb9866

Commits on Mar 20, 2025

• [README] Update notes for global batch size

  yzhangcs

  committed
  3454145

  Copy full SHA for 3454145

• Update conversion path

  yzhangcs

  authored
  56449d7

  Copy full SHA for 56449d7

• Fix isort indentation

  yzhangcs

  committed
  4dc6ffc

  Copy full SHA for 4dc6ffc

• Fix security issues regarding tj-actions

  yzhangcs

  committed
  8108232

  Copy full SHA for 8108232

• Move training configs into model folder

  yzhangcs

  committed
  e61e8d7

  Copy full SHA for e61e8d7

• [Utils] Move conversion scripts into `flame.utils`

  yzhangcs

  committed
  aea1f3c

  Copy full SHA for aea1f3c

Commits on Mar 14, 2025

• [README] Add explanations for some key params

  yzhangcs

  authored
  98517a9

  Copy full SHA for 98517a9

Commits on Mar 13, 2025

• Fix shortconv bugs in non-varlen mode (#19)

  yzhangcs

  committed
  994d4e6

  Copy full SHA for 994d4e6

Pagination

Previous

Next