Commits

Branch selector

 main

User selector

All users

Datepicker

All time

Commit History

Commits on Jul 13, 2025

• fixing dtype in flux eval (#1388)

  Show description for 05e47c3
  kfirgoldberg

  authored
  05e47c3

  Copy full SHA for 05e47c3

Commits on Jul 12, 2025

• [benchmark] add h200 bench (#1361)

  Show description for cbccb38
  asaiacai

  authored
  cbccb38

  Copy full SHA for cbccb38

Commits on Jul 11, 2025

• Fix remaining flake8 and lint in titan repo (#1382)

  Show description for 3ca7041
  H-Huang

  authored
  3ca7041

  Copy full SHA for 3ca7041

• Fix tokenizer README repository ID (#1381)

  Show description for 5375abb
  H-Huang

  authored
  5375abb

  Copy full SHA for 5375abb

• [DSV3] Adding deepseek-v3 model into torchtitan (#1373)

  Show description for d54d05a
  

  3 peopleauthored
  d54d05a

  Copy full SHA for d54d05a

Commits on Jul 10, 2025

• Refactor Tokenizer -> BaseTokenizer (#1333)

  Show description for 681df50
  H-Huang

  authored
  681df50

  Copy full SHA for 681df50

• Migrating flux checkpointing to hf api (#1377)

  Show description for 8518306
  wesleytruong

  authored
  8518306

  Copy full SHA for 8518306

• [SimpleFSDP] Add support for hsdp+tp (#1343)

  Show description for f0ce21b
  ruisizhang123

  authored
  f0ce21b

  Copy full SHA for f0ce21b

• basic validator implementation (#1362)

  Show description for acd5ba8
  wesleytruong

  authored
  acd5ba8

  Copy full SHA for acd5ba8

Commits on Jul 9, 2025

• [float8 moe] Support TP (#1375)

  Show description for be15836
  danielvegamyhre

  authored
  be15836

  Copy full SHA for be15836

Commits on Jul 8, 2025

• call init_weights before generation (#1371)

  Show description for 170b992
  jquesnelle

  authored
  170b992

  Copy full SHA for 170b992

• fix: correct CONFIG_FILE path in multinode_trainer script (#1374)

  idoh

  authored
  5a26243

  Copy full SHA for 5a26243

• dp2ep Expert Parallel (#1324)

  Show description for 01f4e50
  tianyu-l

  authored
  01f4e50

  Copy full SHA for 01f4e50

Commits on Jul 3, 2025

• [WIP][RFC] Always flatten model state_dict (#1347)

  Show description for 7d5f3cc
  fegin

  authored
  7d5f3cc

  Copy full SHA for 7d5f3cc

• Add debugging instructions for "Reproducibility between runs" (#1363)

  Show description for 4aa9fde
  

  wwwjn

  and

  tianyu-l

  authored
  4aa9fde

  Copy full SHA for 4aa9fde

Commits on Jul 2, 2025

• Support different tokenizers (#1318)

  Show description for a04f6bd
  H-Huang

  authored
  a04f6bd

  Copy full SHA for a04f6bd

• [float8 moe training] validate float8 moe parallelism config (#1360)

  Show description for c08c9d4
  danielvegamyhre

  authored
  c08c9d4

  Copy full SHA for c08c9d4

Commits on Jul 1, 2025

• Titan changes to use DCP ZOC instead of titan default Async + Pinned Memory (#1287)

  Show description for c31ff8b
  

  Saiteja64

  and

  saiteja64

  authored
  c31ff8b

  Copy full SHA for c31ff8b

• [float8] add _auto_filter_for_recipe for float8 training (#1319)

  Show description for b0902b2
  danielvegamyhre

  authored
  b0902b2

  Copy full SHA for b0902b2

• [float8 moe training] Do MoE conversion before Float8Linear conversion to avoid unconverting Float8Linears (#1359)

  Show description for 42b0beb
  danielvegamyhre

  authored
  42b0beb

  Copy full SHA for 42b0beb

• Suppress the PyTorch warning when verifying LR Scheduler (#1356)

  Show description for 8004fef
  fegin

  authored
  8004fef

  Copy full SHA for 8004fef

Commits on Jun 27, 2025

• Only calls destroy_process_group if the trainer exist successfully (#1342)

  Show description for 6b11290
  fegin

  authored
  6b11290

  Copy full SHA for 6b11290

• unit test for flux_dataset dataloader checkpointing (#1346)

  Show description for 5d4cc9a
  wesleytruong

  authored
  5d4cc9a

  Copy full SHA for 5d4cc9a

• [llama4] add back max_len arg to generate_permute_indices (#1354)

  Show description for 05d3f7c
  danielvegamyhre

  authored
  05d3f7c

  Copy full SHA for 05d3f7c

• Update the code owners (#1352)

  fegin

  authored
  0c368fe

  Copy full SHA for 0c368fe

• [float8] add float8 rowwise MoE prototype (#1245)

  Show description for 5990fa3
  danielvegamyhre

  authored
  5990fa3

  Copy full SHA for 5990fa3

Commits on Jun 26, 2025

• Always ignore freqs_cis (#1338)

  Show description for aefe15a
  fegin

  authored
  aefe15a

  Copy full SHA for aefe15a

• missing dependency in pyproject for tyro (#1335)

  Show description for 71b07ad
  

  wesleytruong

  and

  wesleytruong

  authored
  71b07ad

  Copy full SHA for 71b07ad

• Update .gitignore to make it friendly to some users (#1337)

  Show description for e0b4534
  fegin

  authored
  e0b4534

  Copy full SHA for e0b4534

• Log `grad_norm` (#1339)

  Show description for a85e2a5
  runame

  authored
  a85e2a5

  Copy full SHA for a85e2a5

• Revert "[kernels] move generate_permute_indices to using exact sizes per expert needed, instead of max_len" (#1340)

  Show description for dc7fd23
  tianyu-l

  authored
  dc7fd23

  Copy full SHA for dc7fd23

Commits on Jun 25, 2025

• [RFC] Make last_save_model_weights_only default to True (#1336)

  Show description for adae6b6
  fegin

  authored
  adae6b6

  Copy full SHA for adae6b6

Commits on Jun 24, 2025

• [CI] Only run integration test when PR's base branch is main (#1331)

  Show description for d9cc6b4
  wwwjn

  authored
  d9cc6b4

  Copy full SHA for d9cc6b4

• Add check for `seq_len%tensor_parallel_degree==0` for parallelized Llama (#1312)

  Show description for 6f6a897
  

  jc-audet

  and

  tianyu-l

  authored
  6f6a897

  Copy full SHA for 6f6a897

Commits on Jun 23, 2025

• [deepseek][blackwell] add manual looping group gemm to enable base working inference on Blackwell (#1272)

  Show description for f234292
  lessw2020

  authored
  f234292

  Copy full SHA for f234292

Pagination

Previous

Next