The collect-results.yml workflow generates tput_vs_intvty and tput_vs_e2el performance graphs but only uploads them as artifacts, requiring users to download files to view results.

Changes

• New utility: utils/display_graphs.py

  • Finds generated PNG files matching tput_vs_intvty_*_{exp_name}.png and tput_vs_e2el_*_{exp_name}.png
  • Generates markdown links to the GitHub Actions artifacts page
  • Groups graphs by type (interactivity vs e2e latency) and model family
  • Accepts repository info, run ID, and artifact name to construct proper URLs

• Workflow update: .github/workflows/collect-results.yml

  • Added step after artifact upload to display graph links in $GITHUB_STEP_SUMMARY
  • Passes GitHub context variables (github.repository, github.run_id) to the script
  • Preserves existing artifact upload for archival

Result

All workflows calling collect-results.yml now display links to performance graphs in the Actions summary:

## Performance Graphs

### Throughput vs Interactivity
#### 70B
📊 [tput_vs_intvty_70b_all.png](https://github.com/InferenceMAX/InferenceMAX/actions/runs/12345/artifacts) (download `graphs_all` artifact)

### Throughput vs End-to-End Latency
#### 70B
📊 [tput_vs_e2el_70b_all.png](https://github.com/InferenceMAX/InferenceMAX/actions/runs/12345/artifacts) (download `graphs_all` artifact)

This approach provides direct links to the artifacts page with clear download instructions, avoiding issues with large embedded images.

View original Slack conversation

View original Slack conversation

💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more Copilot coding agent tips in the docs.