Skip to content

Conversation

AkazaAkane
Copy link
Contributor

Motivation

Modifications

Updated references to flashinfer_python in:

  • docs/start/install.md
  • python/pyproject.toml
  • python/sglang/srt/entrypoints/engine.py
  • scripts/ci_install_dependency.sh

Update BatchDecodeWithPagedKVCacheWrapper Plan (API Document) and remove cuda stream sync in:

  • python/sglang/srt/layers/attention/flashinfer_backend.py
  • python/sglang/srt/layers/attention/flashinfer_mla_backend.py

Checklist

@AkazaAkane
Copy link
Contributor Author

@sleepcoo Seems to have the same packages version conflict as urs. Any ideas on how to address that?

@sleepcoo
Copy link
Collaborator

@sleepcoo Seems to have the same packages version conflict as urs. Any ideas on how to address that?

Here's the issue flashinfer-ai/flashinfer#911 , the flashinfer API is incompatible with the PyTorch distribution. I'm not sure what the best solution is. @zhyncs , do you have any suggestions?

@merrymercy
Copy link
Contributor

can you fix the conflicts?

@AkazaAkane
Copy link
Contributor Author

can you fix the conflicts?
The error is related to pytorch 2.6. I am able to adopt flashinfer to 0.2.5 with a lower version of torch, but I am not able to do 2.6 right now.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Projects
None yet
Development

Successfully merging this pull request may close these issues.

5 participants