Skip to content

feat(inference): time-travel chart data to the selected run#459

Merged
adibarra merged 1 commit into
masterfrom
fix/data-point-run-link
Jun 14, 2026
Merged

feat(inference): time-travel chart data to the selected run#459
adibarra merged 1 commit into
masterfrom
fix/data-point-run-link

Conversation

@adibarra

@adibarra adibarra commented Jun 14, 2026

Copy link
Copy Markdown
Contributor

Note

Medium Risk
Changes core benchmark fetch and SQL filtering used by the main chart; behavior is covered by new tests but wrong cutoff logic could show stale or missing points.

Overview
The inference chart can now show benchmark data as of a selected GitHub workflow run, not only the latest sweep for a day.

Backend: GET /api/v1/benchmarks accepts an optional numeric runId. SQL and JSON getLatestBenchmarks apply a run cutoff on the main chart’s date-filtered path (run_started_at ≤ selected run), while keeping rows with null timestamps and skipping the filter for exact=true GPU-comparison queries.

Frontend: Run selection is resolved before chart fetch; asOfRunId is sent only when the user picks an earlier run than the latest for the model. That forces the dated query path (not the materialized-view shortcut) and extends React Query keys so per-run views don’t collide in cache.

Reviewed by Cursor Bugbot for commit 8d047d0. Bugbot is set up for automated code reviews on this repo. Configure here.

@adibarra adibarra marked this pull request as ready for review June 14, 2026 07:40
@vercel

vercel Bot commented Jun 14, 2026

Copy link
Copy Markdown

The latest updates on your projects. Learn more about Vercel for GitHub.

Project Deployment Actions Updated (UTC)
inferencemax-app Ready Ready Preview, Comment Jun 14, 2026 7:40am

Request Review

@blacksmith-sh

blacksmith-sh Bot commented Jun 14, 2026

Copy link
Copy Markdown

Found 6 test failures on Blacksmith runners:

Failures

Test View Logs
Model Architecture Diagram › Alternating Attention Blocks (MoE model - gpt-oss 120B)/
"before all" hook for "shows MoE and Sink/Full GQA badges for gpt-oss"
View Logs
Model Architecture Diagram › Alternating Attention Blocks (MoE model - gpt-oss 120B)/
"before all" hook for "shows MoE and Sink/Full GQA badges for gpt-oss"
View Logs
Model Architecture Diagram › Collapsible Transformer Blocks (MoE model - MiniMax M3)/
"before all" hook for "shows MoE and GQA badges for MiniMax M3"
View Logs
Model Architecture Diagram › Collapsible Transformer Blocks (MoE model - MiniMax M3)/
"before all" hook for "shows MoE and GQA badges for MiniMax M3"
View Logs
Model Architecture Diagram › Hybrid Attention Blocks (MoE model - DeepSeek V4 Pro)/
"before all" hook for "shows MoE and Hybrid badges for DeepSeek V4 Pro"
View Logs
Model Architecture Diagram › Hybrid Attention Blocks (MoE model - DeepSeek V4 Pro)/
"before all" hook for "shows MoE and Hybrid badges for DeepSeek V4 Pro"
View Logs

Fix with Codesmith
Need help on this PR? Tag /codesmith with what you need.

@adibarra adibarra merged commit 13038d7 into master Jun 14, 2026
22 of 24 checks passed
@adibarra adibarra deleted the fix/data-point-run-link branch June 14, 2026 07:43
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