Skip to content

ci: fix flashinfer-cubin 0.6.14 install by adding flashinfer.ai extra index - #5258

Merged
tzhouam merged 1 commit into
dev/vllm-alignfrom
fix/flashinfer-cubin-extra-index
Jul 21, 2026
Merged

tzhouam merged 1 commit into
dev/vllm-alignfrom
fix/flashinfer-cubin-extra-index

Conversation

@tzhouam

@tzhouam tzhouam commented Jul 21, 2026

Copy link
Copy Markdown
Collaborator

Problem

Nightly align image builds fail at the flashinfer install layer of docker/Dockerfile.ci (first surfaced in build 2803):

× No solution found when resolving dependencies:
╰─▶ Because there is no version of flashinfer-cubin==0.6.14 ...

flashinfer-cubin stopped publishing to PyPI at 0.6.14 — its latest PyPI release is 0.6.13. The 2026-07-18 pin sync (8508bf5) copied upstream's 0.6.14 version bump but not the index change that came with it: upstream vLLM (ba47bb5be, vllm-project/vllm#47669, now on releases/v0.26.0) installs cubin via --extra-index-url https://flashinfer.ai/whl/, where the 0.6.14 wheel exists.

The July 18–20 nightlies never reached the image step (they failed earlier at Load pipeline), which is why this only surfaced on 7/21.

Fix

Add --extra-index-url https://flashinfer.ai/whl/ to the cubin/python install layer, matching upstream. Pins stay at 0.6.14.

Verification

uv pip install --dry-run "flashinfer-python==0.6.14" "flashinfer-cubin==0.6.14" --extra-index-url https://flashinfer.ai/whl/ resolves both (cubin from flashinfer.ai, python from PyPI). The flashinfer-jit-cache==0.6.14 layer below already uses the cu130 index and is unaffected.

🤖 Generated with Claude Code

… index

flashinfer-cubin stopped publishing to PyPI at 0.6.14; upstream vLLM pulls
it via --extra-index-url https://flashinfer.ai/whl/ (ba47bb5be, PR #47669).
The 2026-07-18 pin sync copied the 0.6.14 version bump but not the index,
so the nightly image build fails at this layer with 'no version of
flashinfer-cubin==0.6.14' (first surfaced in build 2803; earlier nightlies
died before reaching the image step).

Resolution verified with uv dry-run: both pins resolve with the extra index.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
@chatgpt-codex-connector

Copy link
Copy Markdown

Codex usage limits have been reached for code reviews. Please check with the admins of this repo to increase the limits by adding credits.
Credits must be used to enable repository wide code reviews.

@tzhouam
tzhouam merged commit 5ec35f9 into dev/vllm-align Jul 21, 2026
0 of 3 checks passed
@tzhouam
tzhouam deleted the fix/flashinfer-cubin-extra-index branch July 21, 2026 06:12
tzhouam added a commit that referenced this pull request Jul 23, 2026
… index (#5258)

Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
tzhouam added a commit that referenced this pull request Jul 27, 2026
… index (#5258)

Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
tzhouam added a commit that referenced this pull request Jul 27, 2026
… index (#5258)

Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
Claisenn added a commit to Claisenn/vllm-omni that referenced this pull request Aug 1, 2026
… index (vllm-project#5258)

Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant