[amd_dev] branch rebase #25753
[amd_dev] branch rebase #25753
Conversation
Signed-off-by: Nick Hill <[email protected]>
Signed-off-by: Lucas Kabela <[email protected]>
Signed-off-by: Max de Bayser <[email protected]>
…Prompt Embeds support (#25291) Signed-off-by: Andrew Sansom <[email protected]>
Signed-off-by: Andrew Sansom <[email protected]>
Signed-off-by: Boyuan Feng <[email protected]> Signed-off-by: Boyuan Feng <[email protected]> Signed-off-by: boyuanfeng <[email protected]> Co-authored-by: Luka Govedič <[email protected]>
Signed-off-by: Nick Hill <[email protected]> Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
Signed-off-by: Nick Hill <[email protected]>
…5289) Signed-off-by: Harry Mellor <[email protected]>
…utoGPTQ and AutoRound-GPTQ) (#25268) Signed-off-by: JartX <[email protected]>
Signed-off-by: Nick Hill <[email protected]>
…ion (#25298) Signed-off-by: Chendi Xue <[email protected]>
Signed-off-by: chaunceyjiang <[email protected]> Co-authored-by: xin.li <[email protected]>
…ng models (#25261) Signed-off-by: DarkLight1337 <[email protected]>
…25101) Signed-off-by: Chen Zhang <[email protected]>
Signed-off-by: Roger Wang <[email protected]>
Signed-off-by: DarkLight1337 <[email protected]>
Signed-off-by: mgoin <[email protected]>
Signed-off-by: wwl2755 <[email protected]>
…speed (#23558) Signed-off-by: Manoel Marques <[email protected]> Signed-off-by: Manoel Marques <[email protected]> Co-authored-by: Harry Mellor <[email protected]> Co-authored-by: Luka Govedič <[email protected]>
Signed-off-by: Isotr0py <[email protected]>
…erage (#25308) Signed-off-by: pengdrumli <[email protected]>
Signed-off-by: DarkLight1337 <[email protected]>
) Signed-off-by: windsonsea <[email protected]>
Signed-off-by: Woosuk Kwon <[email protected]> Signed-off-by: Woosuk Kwon <[email protected]>
Signed-off-by: Woosuk Kwon <[email protected]> Signed-off-by: Woosuk Kwon <[email protected]>
Signed-off-by: Woosuk Kwon <[email protected]>
Signed-off-by: Woosuk Kwon <[email protected]>
Signed-off-by: Woosuk Kwon <[email protected]> Signed-off-by: Woosuk Kwon <[email protected]>
Signed-off-by: Huamin Li <[email protected]> Co-authored-by: Lu Fang <[email protected]>
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request appears to be a large rebase of a development branch, introducing a wide array of changes across the codebase. Key updates include a major refactoring of the CI/CD pipelines, deprecation of old benchmark scripts in favor of a new CLI, and the addition of numerous new benchmarks. On the feature side, there's new support for FP8 on ROCm, various new fused kernels for performance, and significant improvements to the CPU backend with oneDNN and scalar fallbacks. I've identified two critical correctness issues related to potential out-of-bounds memory access in MoE kernels, which I've detailed in the comments below. The rest of the changes, including new features and extensive refactoring, appear solid.
| int expert_id = topk_ids[i]; | ||
| if (expert_id >= num_experts) { | ||
| continue; | ||
| } |
There was a problem hiding this comment.
This bounds check on expert_id is a critical fix. Without it, an invalid ID from topk_ids could lead to an out-of-bounds atomic add on shared_counts, potentially causing memory corruption or a crash. This is an important safeguard for correctness and stability.
| int32_t expert_id = topk_ids[i]; | ||
| if (expert_id >= num_experts) { | ||
| continue; | ||
| } |
There was a problem hiding this comment.
Adding this bounds check for expert_id is crucial for correctness and security. It prevents potential out-of-bounds writes to cumsum_buffer and sorted_token_ids if topk_ids contains an invalid expert ID, which could otherwise lead to memory corruption or crashes.
![github-advanced-security[bot]](https://avatars.githubusercontent.com/in/57789?s=60&v=4){kind=small}
| if ctype is None: | ||
| ctype = {"jpg": "image/jpg", "png": "image/png"}[ext] | ||
| self.send_response(200) | ||
| self.send_header("Content-Type", ctype) |
Check warning
Code scanning / CodeQL
HTTP Response Splitting Medium test
Show autofix suggestion
Hide autofix suggestion
Copilot Autofix
AI 4 days ago
To fix the potential HTTP response splitting vulnerability, sanitize or strictly validate any user-derived input used for constructing HTTP header values. Here, before using ctype as a header value, we should ensure that it contains no CR, LF, or colon characters that could allow header injection. We can do this by stripping or replacing these characters if present. It's also prudent to be defensive, as a belt-and-braces approach; even though in this code only jpg/png extensions are accepted, we should sanitize just in case. The fix will:
- Strip/disallow
\r,\n, and:characters fromctypebefore it is used insend_header. - Do this immediately before using
ctypeon line 1173.
No new methods or imports are necessary, as simple string methods suffice for this sanitization.
| @@ -1170,7 +1170,8 @@ | ||
| if ctype is None: | ||
| ctype = {"jpg": "image/jpg", "png": "image/png"}[ext] | ||
| self.send_response(200) | ||
| self.send_header("Content-Type", ctype) | ||
| safe_ctype = ctype.replace("\n", "").replace("\r", "").replace(":", "") | ||
| self.send_header("Content-Type", safe_ctype) | ||
| self.send_header("Content-Length", str(len(data))) | ||
| self.end_headers() | ||
| self.wfile.write(data) |
…#25698) Signed-off-by: Sage Moore <[email protected]> Co-authored-by: Robert Shaw <[email protected]>
Signed-off-by: 许文卿 <[email protected]>
Signed-off-by: DarkLight1337 <[email protected]>
Signed-off-by: Chih-Chieh-Yang <[email protected]> Co-authored-by: RishiAstra <[email protected]>
Signed-off-by: chaunceyjiang <[email protected]>
Signed-off-by: wang.yuqi <[email protected]> Co-authored-by: Cyrus Leung <[email protected]>
Signed-off-by: DarkLight1337 <[email protected]>
Signed-off-by: DarkLight1337 <[email protected]>
github-project-automation
bot
moved this from To Triage
to In progress
in gpt-oss Issues & Enhancements
|
@yewentao256 try to land it. |
github-project-automation
bot
moved this from In progress
to Ready
in gpt-oss Issues & Enhancements
…25455) Signed-off-by: Isotr0py <[email protected]>
HAIAI
added
the
ready-for-merge
github-project-automation
bot
moved this from Ready
to Done
in gpt-oss Issues & Enhancements
Purpose
rebase/sync branch to latest main
Test Plan
Test Result
Essential Elements of an Effective PR Description Checklist
supported_models.mdandexamplesfor a new model.