Skip to content

sglang: bump to 0.5.19 + DeepEP v2 over EFA (amzn2023) - #6745

Draft
jinyan-li1 wants to merge 6 commits into
mainfrom
deepep-v2/sglang-amzn2023
Draft

jinyan-li1 wants to merge 6 commits into
mainfrom
deepep-v2/sglang-amzn2023

Conversation

@jinyan-li1

Copy link
Copy Markdown
Contributor

Bumps the SGLang amzn2023 image to sglang 0.5.19 and integrates DeepEP v2 over EFA, mirroring the vLLM and vLLM-Omni amzn2023 images.

  • sglang 0.5.17 → 0.5.19 (torch 2.13, sgl-kernel 0.4.6.post1, flashinfer 0.6.18, mooncake 0.3.13, cutlass-dsl 4.6.2)
  • DeepEP v2 from the amazon-contributing/DeepEP fork; NCCL 2.31.2 (cu13), runtime-pinned with a version assert
  • EFA 1.50.0, GDRCopy 2.5.2
  • Prebuild all_reduce_perf so the EFA test setup is a no-op
  • DeepEP-v2 sanity check (deep_ep present + runtime NCCL >= 2.30.4)
  • dlc_minor_version 3 → 4

Draft — validating in CI.

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.

1 participant