Skip to content
Merged
Changes from 3 commits
Commits
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 2 additions & 2 deletions .github/workflows/ci.yml
Original file line number Diff line number Diff line change
Expand Up @@ -91,7 +91,7 @@ jobs:
- tags: tilelang-metax-runner
name: self-hosted-metax
# Format: MACA-<major>.<minor>[.<patch>]. E.g., "MACA-3.0".
toolkit: MACA-3.6
toolkit: MACA-3.7
- tags: [macos-latest]
name: macos-latest
toolkit: Metal # or Nightly-Metal
Expand Down Expand Up @@ -288,7 +288,7 @@ jobs:
uv pip install -v -r requirements-test-rocm.txt
elif [[ "${{ matrix.runner.toolkit }}" == *"MACA"* ]]; then
uv pip install -v -r requirements-test-maca.txt
uv pip install -v --no-deps --python-version 3.10.0 flash_linear_attention==0.4.0+metax3.5.3.9torch2.8 -i https://repos.metax-tech.com/r/maca-pypi/simple --trusted-host repos.metax-tech.com
uv pip install -v --no-deps --python-version 3.10.0 flash_linear_attention==0.4.0 -i https://repos.metax-tech.com/r/maca-pypi/simple --trusted-host repos.metax-tech.com

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

⚠️ Potential issue | 🟠 Major

🧩 Analysis chain

🏁 Script executed:

#!/bin/bash
# Description: Check available flash_linear_attention versions in the MACA repository

# Query the repository for available versions
curl -s "https://repos.metax-tech.com/r/maca-pypi/simple/flash-linear-attention/" 2>/dev/null | \
  grep -oP 'flash[_-]linear[_-]attention-[^"]+\.whl' | \
  grep '0\.4\.0' | \
  head -20

# If curl fails, note that manual verification is needed
if [ $? -ne 0 ]; then
  echo "Note: Unable to query repository. Manual verification required."
fi

Repository: tile-ai/tilelang-metax

Length of output: 1654


🏁 Script executed:

#!/bin/bash
set -euo pipefail

base="https://repos.metax-tech.com/r/maca-pypi/simple/flash-linear-attention/"
page="$(curl -fsSL "$base")"

echo "== wheels matching exact version 0.4.0 (no + local segment) =="
# Example wheel names: flash_linear_attention-0.4.0-cp310-...whl
echo "$page" | rg -o 'flash[_-]linear[_-]attention-0\.4\.0-[^"]+\.whl' | sort -u | head -50

echo
echo "== wheels matching 0.4.0+metax3.7 local versions =="
echo "$page" | rg -o 'flash[_-]linear[_-]attention-0\.4\.0\+metax3\.7[^"]+\.whl' | sort -u | head -50

echo
echo "== any exact 0.4.0 references on the page =="
echo "$page" | rg -n 'flash[_-]linear[_-]attention-0\.4\.0' | head -50

Repository: tile-ai/tilelang-metax

Length of output: 129


🏁 Script executed:

#!/bin/bash
set -uo pipefail

base="https://repos.metax-tech.com/r/maca-pypi/simple/flash-linear-attention/"
page="$(curl -fsSL "$base" 2>/dev/null || true)"

if [[ -z "${page}" ]]; then
  echo "curl returned empty page (or failed)."
  exit 0
fi

echo "== Sample matches from index page =="
echo "$page" | rg -o 'flash[_-]linear[_-]attention-[^"]+\.whl' | head -50

echo
echo "== Unique versions from wheel filenames =="
echo "$page" | rg -o 'flash[_-]linear[_-]attention-([^"]+)\.whl' | sed -E 's/flash[_-]linear[_-]attention-//' | sort -u

echo
echo "== Any exact 0.4.0 (public) wheel link? =="
echo "$page" | rg -n 'flash[_-]linear[_-]attention-0\.4\.0[^+./"]*\.whl' || true

Repository: tile-ai/tilelang-metax

Length of output: 4408


Fix flash_linear_attention pin for MACA 3.7 (avoid installing non-existent public 0.4.0)

The metax index for flash-linear-attention provides wheels only for local versions like flash_linear_attention-0.4.0+metax3.7.0.7torch2.8 / 0.4.0+metax3.7.1.3torch2.8 / 0.4.0+metax3.7.2.0torch2.8 (and torch2.6 variants), and shows no wheel entries for the plain public flash_linear_attention-0.4.0 (without the +metax... suffix). Pinning to flash_linear_attention==0.4.0 at .github/workflows/ci.yml:291 may fail to resolve.

Update the requirement to the appropriate flash_linear_attention==0.4.0+metax3.7...torch... variant matching your MACA 3.7 toolkit and torch version.

🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In @.github/workflows/ci.yml at line 291, The pip install line currently pins
flash_linear_attention to the plain public version
"flash_linear_attention==0.4.0" which doesn't exist on the metax index; update
that package spec in the uv pip install command to the metax-specific wheel
variant (e.g. "flash_linear_attention==0.4.0+metax3.7.X.YtorchZ.W") that matches
your MACA 3.7 toolkit and Torch version so pip can resolve it, keeping the same
index URL and --trusted-host flags.

elif [[ "${{ matrix.runner.toolkit }}" == *"Metal"* ]]; then
uv pip install -v -r requirements-test-metal.txt
else
Expand Down
Loading