Skip to content

Pull requests: pytorch/torchtitan

New pull request New
289 Open 2,311 Closed
289 Open 2,311 Closed
Author
Filter by author
Loading
Label
Filter by label
Loading
Use alt + click/return to exclude labels
or ⇧ + click/return for logical OR
Projects
Filter by project
Loading
Milestones
Filter by milestone
Loading
Reviews
Filter by reviews
No reviews Review required Approved review Changes requested
Assignee
Filter by who’s assigned
Assigned to nobody Loading
Sort
Sort by
Newest Oldest Most commented Least commented Recently updated Least recently updated Best match
Most reactions
👍 👎 😄 🎉 😕 ❤️ 🚀 👀

Pull requests list

[rl] Add optional chunked logprobs to reduce peak memory ciflow/rl ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3314 opened May 11, 2026 by DamianSzwichtenberg Loading…
1
[graph_trainer] Add log_timer utility for tracing step timing ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3311 opened May 8, 2026 by SherlockNoMad Contributor Loading…
1 of 2 tasks
Remove MoE expert for-loop fallback ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3308 opened May 8, 2026 by sanketpurandare Contributor Loading…
2
Revert "[graph_trainer] Pass global_valid_tokens into loss_fn instead of dividing externally" to fix ci ci-no-td ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3307 opened May 8, 2026 by ydwu4 Contributor Loading…
1
[GraphTrainer] Add Context Parallel support ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3305 opened May 8, 2026 by aditvenk Contributor Loading…
1
[RFC] Multi-turn chat dataset CLA Signed This label is managed by the Meta Open Source bot.
#3304 opened May 8, 2026 by felipemello1 Contributor • Draft
4
[draft][spmd_llama3] custom LP ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3302 opened May 8, 2026 by pianpwk Contributor • Draft
[graph_trainer] Fix cpuoff init hang ciflow/rl ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3281 opened May 8, 2026 by IvanKobzarev Contributor • Draft
1
[draft][spmd_llama3] ChunkedCELoss local_map refactor ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3278 opened May 8, 2026 by pianpwk Contributor • Draft
[draft][spmd_llama3] CP ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3277 opened May 8, 2026 by pianpwk Contributor • Draft
[graph_trainer] Add chunked loss bitwise deterministic coverage ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3276 opened May 7, 2026 by ydwu4 Contributor Loading…
2
[graph_trainer] Refactor selective activation remat to in-place ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3271 opened May 7, 2026 by tugsbayasgalan Contributor Loading…
[graph_trainer] Refactor selective activation remat to in-place ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3270 opened May 7, 2026 by tugsbayasgalan Contributor Loading…
23
[optimizer] support mixed optimizers ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3269 opened May 7, 2026 by shuhuayu Contributor Loading…
4
chore(ci): migrate ROCm matrix from 7.1 to 7.2 ciflow/rocm ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot. module: rocm
#3267 opened May 7, 2026 by yuxinwan-amd • Draft
3
[Quantization] Add NVFP4 quantization support using TransformerEngine ciflow/8gpu
#3266 opened May 7, 2026 by tbqh • Draft
2
[Quantization] Add QAT (quantization-aware training) converter ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3265 opened May 7, 2026 by mori360 Contributor Loading…
[Checkpoint] BaseModel state dict pipeline and pure I/O CheckpointManager ciflow/rl ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3263 opened May 7, 2026 by mori360 Contributor Loading…
3
[rl] Move RL H100 CI to A10G ciflow/rl ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3259 opened May 7, 2026 by wwwjn Contributor Loading…
[draft][spmd_llama3] TP+CP ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3255 opened May 6, 2026 by pianpwk Contributor • Draft
[draft][spmd_llama3] FSDP/HSDP ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3254 opened May 6, 2026 by pianpwk Contributor • Draft
[draft][spmd_llama3] ParallelDims, ShardingConfig, parallelize_spmd ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3253 opened May 6, 2026 by pianpwk Contributor • Draft
[loss] Chain-rule grad_output through ChunkedCELoss autograd Function backwards ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3250 opened May 6, 2026 by ydwu4 Contributor Loading…
9
Make ChunkedCELoss support torch.autograd.grad ciflow/h100.8 Trigger H100.8 CI ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3249 opened May 6, 2026 by ydwu4 Contributor Loading…
12
[graph_trainer] Enable ChunkedCELoss for deepseek_v3 and qwen3 ciflow/h100.8 Trigger H100.8 CI ciflow/8gpu CLA Signed This label is managed by the Meta Open Source bot.
#3248 opened May 6, 2026 by ydwu4 Contributor Loading…
3
ProTip! Find all pull requests that aren't related to any open issues with -linked:issue.