Commits

Branch selector

 main

User selector

All users

All time

Commit History

Commits on Jan 6, 2026

• [GPT-OSS] Graduate from experiments to main (#2203)

  Show description for 795a7a0
  shuhuayu

  authored
  795a7a0

  Copy full SHA for 795a7a0

• [docs] Fix missing --model.flavor flags in compiler_toolkit README (#2168) (#2201)

  Show description for e16af85
  BryanBradfo

  authored
  e16af85

  Copy full SHA for e16af85

Commits on Dec 30, 2025

• Add docs to explain COMM_MODE (#2162)

  Show description for 7e4ab85
  fegin

  authored
  7e4ab85

  Copy full SHA for 7e4ab85

Commits on Dec 29, 2025

• make get tp mesh optional in llama4 parallelize (#2185)

  Show description for 62f5806
  danielvegamyhre

  authored
  62f5806

  Copy full SHA for 62f5806

• add attention scaling to varlen for qwen3 (#2178)

  Show description for 5dd9f4c
  liangel-02

  authored
  5dd9f4c

  Copy full SHA for 5dd9f4c

Commits on Dec 26, 2025

• Remove psutil as part of requirements (#2181)

  wwwjn

  authored
  81af883

  Copy full SHA for 81af883

• Bump torchtitan version to v0.2.1 (#2180)

  Show description for 64b5e15
  wwwjn

  authored
  64b5e15

  Copy full SHA for 64b5e15

• add safety checks for varlen (#2179)

  Show description for 5077be6
  liangel-02

  authored
  5077be6

  Copy full SHA for 5077be6

• Support TP when using vLLM engine to run inference w/ torchtitan model definition (#2165)

  Show description for a95d203
  wwwjn

  authored
  a95d203

  Copy full SHA for a95d203

Commits on Dec 24, 2025

• [RL] Support Trainer and Generator Unified Model (#2174)

  Show description for 30ab580
  acisseJZhong

  authored
  30ab580

  Copy full SHA for 30ab580

• Add rocm support for models, flux & torchft integration tests. (#2172)

  Show description for a452121
  akashveramd

  authored
  a452121

  Copy full SHA for a452121

Commits on Dec 23, 2025

• Fix qwen3 attention scaling calculation (#2173)

  Show description for 29aafb9
  wwwjn

  authored
  29aafb9

  Copy full SHA for 29aafb9

Commits on Dec 22, 2025

• Multiprocess simple RL loop (#2158)

  Show description for 4b3d25a
  acisseJZhong

  authored
  4b3d25a

  Copy full SHA for 4b3d25a

Commits on Dec 19, 2025

• Bump tj-actions/changed-files from d6e91a2266cdb9d62096cebf1e8546899c6aa18f to e0021407031f5be11a464abee9a0776171c79891 in the github-actions group (#2167)

  Show description for 1bd2548
  dependabot[bot]

  authored
  1bd2548

  Copy full SHA for 1bd2548

• Add Dependabot for GitHub Actions updates (#2163)

  Show description for b21555f
  salmanmkc

  authored
  b21555f

  Copy full SHA for b21555f

• Replace `logger.warn()` to `logger.warning()` , allow `log_validation` to log `extra_metrics` and expose common wandb args (#2166)

  Show description for b786a3d
  EquationWalker

  authored
  b786a3d

  Copy full SHA for b786a3d

• Expose common dataloader args (#2097)

  Show description for 658f94c
  divyanshk

  authored
  658f94c

  Copy full SHA for 658f94c

• Upgrade GitHub Actions for Node 24 compatibility (#2164)

  Show description for fd49b4b
  salmanmkc

  authored
  fd49b4b

  Copy full SHA for fd49b4b

• Fix pypa/gh-action-pypi-publish version to use SHA pinning (#2161)

  Show description for 4438764
  salmanmkc

  authored
  4438764

  Copy full SHA for 4438764

Commits on Dec 18, 2025

• Integrate DeepEP to torchtitan (#2107)

  Show description for 36a4b69
  elfiegg

  authored
  36a4b69

  Copy full SHA for 36a4b69

Commits on Dec 17, 2025

• Use new DeviceMesh unflatten to rewrite parallel_dims (#1660)

  Show description for 183a0d2
  fegin

  authored
  183a0d2

  Copy full SHA for 183a0d2

Commits on Dec 16, 2025

• [RELAND] Let CUDA and ROCm read different loss result (#2157)

  Show description for f64bbad
  fegin

  authored
  f64bbad

  Copy full SHA for f64bbad

• Run vLLM inference using torchtitan model definition (single GPU) (#2119)

  Show description for e36d027
  wwwjn

  authored
  e36d027

  Copy full SHA for e36d027

Commits on Dec 15, 2025

• Add local built pytorch path for pyrefly (#2155)

  Show description for c08fa57
  fegin

  authored
  c08fa57

  Copy full SHA for c08fa57

• [GPT-OSS] Add HF state dict adapter to support loading from HF checkpoints (#2021)

  Show description for 64997d2
  shuhuayu

  authored
  64997d2

  Copy full SHA for 64997d2

• Improve the loss_compare.sh logic (#2143)

  Show description for c283a84
  fegin

  authored
  c283a84

  Copy full SHA for c283a84

Commits on Dec 14, 2025

• Upgrade GitHub Actions for Node 24 compatibility (#2151)

  Show description for f3748d8
  salmanmkc

  authored
  f3748d8

  Copy full SHA for f3748d8

• Upgrade GitHub Actions to latest versions (#2152)

  Show description for c1f4e94
  salmanmkc

  authored
  c1f4e94

  Copy full SHA for c1f4e94

• [MoE] Add node limited routing support (#2111)

  Show description for 2aac20a
  shuhuayu

  authored
  2aac20a

  Copy full SHA for 2aac20a

Commits on Dec 13, 2025

• Implement ciflow/rocm on Torchtitan (#2114)

  Show description for 9bc50ea
  

  akashveramd

  and

  huydhn

  authored
  9bc50ea

  Copy full SHA for 9bc50ea

• [Autoparallel] Add local_map variant of DSv3 and 2D mesh AP (#2129)

  Show description for 995154f
  xmfan

  authored
  995154f

  Copy full SHA for 995154f

Commits on Dec 12, 2025

• Enable static type checking with Pyrefly (#2136)

  Show description for 64dc922
  rchen152

  authored
  64dc922

  Copy full SHA for 64dc922

• Fix apply_compile called multiple times in PP initialization (#2135)

  Show description for 7a398ea
  xmfan

  authored
  7a398ea

  Copy full SHA for 7a398ea

• Enable PP and EP overlap for MoE (#1721)

  Show description for fcc5643
  

  H-Huang

  and

  tianyu-l

  authored
  fcc5643

  Copy full SHA for fcc5643

• Clarify contribution guidelines. (#2134)

  dcci

  authored
  669845f

  Copy full SHA for 669845f