Lightning-AI / Lightning-AI/lightning-thunder

Investigate `torch.Tensor.__setitem__` as unsupported in `thunderfx`

Open
#2,790 3 comments 0 reactions 2 assignees View on GitHub

@shino16 is already working on this.

Since Dec 9, 2025.

operators
Dominant language
Python
Stars
1.5k
Forks
121
PR merge metrics
No merged PRs in 30d

Description

Setting this operator in https://github.com/Lightning-AI/lightning-thunder/blob/f6f0496cc5f0c9071b62ef79a72ffed1a1bcb296/thunder/dynamo/utils.py#L48 gives 10x decode throughput improvement on `openai/gpt-oss-20b` and `Qwen/Qwen3-32B` inside SGLang `bench_one_batch.py`.

Repro (this only works on the NV internal [thunder-sglang-integration](https://gitlab-master.nvidia.com/kaiximatteoc/sglang/-/tree/thunder-dynamo-recompilation) codebase):
```
SGLANG_USE_THUNDER_GRAPH_RUNNER=1 python3 -m sglang.bench_one_batch --model-path openai/gpt-oss-20b --trust-remote-code --model-impl transformers --dtype bfloat16 --json-model-override-args '{"quantization_config": null}' --cuda-graph-bs 1 --tp-size 4 --tp-strategy dtensor --load-format-dummy
```

Contributor guide

No contributing guide indexed for this repository

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.