[Ascend]Support qwen3.5#18544
Conversation
Summary of ChangesHello @chenxu214, I'm Gemini Code Assist1! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed! This pull request introduces initial support for the Qwen3.5 model, specifically targeting Ascend NPU environments. The changes involve adapting the attention backend to recognize and operate correctly on NPUs, refining the quantization process to handle model-specific prefixing, and updating the Qwen3.5 model's internal component naming for MLP layers to ensure accurate configuration. Highlights
🧠 New Feature in Public Preview: You can now enable Memory to help Gemini Code Assist learn from your team's feedback. This makes future code reviews more consistent and personalized to your project's style. Click here to enable Memory in your admin console. Changelog
Activity
Using Gemini Code AssistThe full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips. Invoking Gemini You can request assistance from Gemini at any point by creating a comment using either
Customization To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a Limitations & Feedback Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments. If you're interested in giving your feedback about your experience with Gemini Code Assist for Github and other Google products, sign up here. You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension. Footnotes
|
There was a problem hiding this comment.
Code Review
This pull request adds support for qwen3.5 on Ascend NPUs. The changes correctly adjust import paths and logic to be compatible with NPU devices. I've found one area for improvement in python/sglang/srt/layers/quantization/modelslim/modelslim.py where a method has side effects and is inefficient. I've suggested a fix to address this. Other changes look good.
|
/tag-and-rerun-ci |
004523a to
9f15540
Compare
This PR affects only the NPU. If any issues arise, please contact iforgetmyname.
This PR affects only the NPU. If any issues arise, please contact iforgetmyname.
|
We faced a fail with Radix cache. Did you test with enable Radix cache? |
This PR affects only the NPU. If any issues arise, please contact iforgetmyname.
Motivation
Ascend adapts qwen3.5
Modifications
1、Bugfix with load weights： issue
2、The quantization and import kernel error reporting parts have been modified accordingly.
Accuracy Tests
Benchmarking and Profiling
python -m sglang.launch_server
--model-path $MODELPATH
--host ${MIX_IP[$i]} --port 6699
--tp-size 16
--quantization modelslim
--nnodes 1 --node-rank $i
--attention-backend ascend
--mem-fraction-static 0.85
--max-running-requests 16
--watchdog-timeout 3600
--disable-radix-cache
--max-prefill-tokens 16384
--max-total-tokens 120000
--chunked-prefill-size -1
--mamba-ssm-dtype bfloat16
--cuda-graph-bs 1 2 4 8 12 16
--enable-multimodal
--mm-attention-backend ascend_attn
--dtype bfloat16
Checklist
Review Process
/tag-run-ci-label,/rerun-failed-ci,/tag-and-rerun-ci