🚀 A simple way to launch, train, and use PyTorch models on almost any device and distributed configuration, automatic mixed precision (including fp8), and easy-to-configure FSDP and DeepSpeed support
active 2025-03-06 → 2026-07-08 (UTC)
Activity over time
Daily event counts in the loaded window
Line chart, 490 days from 2025-03-06 to 2026-07-08. Pushes: 240 total, peak 15 in a day. Pull requests: 228 total, peak 9 in a day. Issues: 211 total, peak 5 in a day. Comments: 825 total, peak 19 in a day. Stars: 567 total, peak 6 in a day.
- Pushes
- Pull requests
- Issues
- Comments
- Stars
Stars, PRs, issues and forks are under-captured in the later part of this window. GH Archive progressively stopped capturing non-push events during 2026 — −95% or worse by the end of the window. Every series here except Pushes fades for that reason, so a decline above reflects the archive, not this repository. Pushes stay reliable throughout, so read them, and the contributor counts derived from them, as the real signal. Data health has the measurements.
Top contributors
Pushes, PRs, issues, reviews and comments — stars and forks excluded, so this is contribution rather than popularity
| Contributor | Contributions | Pushes | PRs | Comments |
|---|---|---|---|---|
| SunMarc | 411 | 121 | 64 | 108 |
| S1ro1 | 230 | 66 | 25 | 89 |
| github-actions[bot] | 200 | 1 | 13 | 111 |
| HuggingFaceDocBuilderDev | 73 | 0 | 0 | 73 |
| IlyasMoutawwakil | 69 | 32 | 6 | 17 |
| yao-matrix | 62 | 0 | 13 | 38 |
| stas00 | 29 | 0 | 1 | 19 |
| SalmanMohammadi | 25 | 0 | 1 | 12 |
| kashif | 21 | 0 | 3 | 11 |
| kmehant | 14 | 0 | 3 | 10 |
| cyr0930 | 11 | 0 | 0 | 6 |
| naomili0924 | 11 | 0 | 1 | 9 |
| sayakpaul | 11 | 5 | 0 | 4 |
| pcuenca | 10 | 4 | 0 | 5 |
| jiqing-feng | 10 | 0 | 1 | 4 |
| winglian | 10 | 0 | 2 | 4 |
| djsaunde | 9 | 0 | 0 | 5 |
| xliu0105 | 9 | 0 | 1 | 6 |
| ldh127 | 8 | 0 | 0 | 7 |
| ojh31 | 8 | 0 | 2 | 3 |
Recent activity
Latest issues, pull requests and releases
- Pull request#4085dependabot[bot]2026-06-29 07:54
- Issue#4075GoldenStain2026-06-12 10:39[Feature request] Support already-sharded DataLoaders in Accelerator.prepare
- Pull request#4065lollinng2026-06-09 15:11
- Pull request#4059SunMarc2026-05-29 14:06
- Issue#3991github-actions[bot]2026-05-27 16:21Isssue when using torch.compile
- Issue comment#4017yuxinyuan2026-05-21 02:38fix: compile inner model before DDP wrapping to prevent Dynamo tracing DDP internals
- Issue#4035JesesePU2026-05-09 22:46Anyone seen TSU Protocol? Anonymous open-source chip design
- Issue#4033JesesePU2026-05-09 21:21Sharing: TSU Protocol — open-source AI chips without a company
- Issue#4029loongmiaow-pixel2026-05-05 03:54[Windows] RTX 5070 Ti (Blackwell sm_120) - setup and deployment notes
- Issue#3990github-actions[bot]2026-05-04 15:48[Feature] Save model-only feature in `save_state`
- Issue comment#3995github-actions[bot]2026-05-04 15:48device_map="auto": silent corruption of tensors captured in register_forward_hook (3+ GPUs, inference_mode, stale bare references)
- Pull request#4027imstevenpmwork2026-05-03 13:10
- Issue#3979github-actions[bot]2026-04-26 15:20[Bug] FSDP2 mixed-precision upcast to fp32 is a silent no-op since v1.13.0
- Issue#4016AlliedToasters2026-04-22 23:21Feature request: disk_offload() that mmaps HF-cache safetensors directly (no duplicate copy, no CPU-RAM materialization)
- Issue comment#3985SunMarc2026-04-21 12:49[FSDP2] Cast model to uniform dtype before fully_shard to fix mixed-dtype AssertionError
- Issue comment#4010dercodeKoenig2026-04-18 01:32Ram explodes when training a very simple model with accelerate tpu
- Pull request#4008rtrompier2026-04-14 11:54
- Issue comment#3708livehappy12026-04-14 03:05Detected kernel version 5.4.250, which is below the recommended minimum of 5.5.0;
- Issue#3956github-actions[bot]2026-04-12 15:18`infer_auto_device_map` does not place submodule buffers on `device_map` when submodule is split
- Issue#3954github-actions[bot]2026-04-12 15:18AssertionError for FP8 Benchmarks
- Pull request#3987roycho962026-04-10 14:13
- Issue comment#4000czkkkkkk2026-04-09 21:31Add padded allgather and broadcast for Neuron devices to reduce recompilation
- Pull request#3970liuyun73452026-04-09 15:49
- Pull request#4003hf-security-analysis[bot]2026-04-08 14:30
- Pull request#4002hf-security-analysis[bot]2026-04-08 14:29
Totals cover only the window loaded into ClickHouse and count events, not GitHub's lifetime totals — 567 stars here means stars gained during the window, not the repo's star count.