test: validate MLX inference and scoring pipeline on M3 Ultra #2

New issue

Closed

opened 2026-02-16 17:16:23 +00:00 by clotho · 1 comment

clotho commented

2026-02-16 17:16:23 +00:00

Member

Run the go-ai test suite on the M3 Ultra with native Metal acceleration.

Build and run all tests: go test ./...
If MLX backend is available, verify ml serve starts and responds to probe requests
Check that scoring (heuristic + judge) works end-to-end
Verify model loading from /Volumes/Data/lem/gguf/ if GGUF files are present
Report test results, any failures, and hardware-specific issues

This runs on the M3 Ultra (80 GPU cores, 96GB) — ideal for validating Metal/MLX paths.

Post findings as a comment on this issue.

Run the go-ai test suite on the M3 Ultra with native Metal acceleration. 1. Build and run all tests: `go test ./...` 2. If MLX backend is available, verify `ml serve` starts and responds to probe requests 3. Check that scoring (heuristic + judge) works end-to-end 4. Verify model loading from `/Volumes/Data/lem/gguf/` if GGUF files are present 5. Report test results, any failures, and hardware-specific issues This runs on the M3 Ultra (80 GPU cores, 96GB) — ideal for validating Metal/MLX paths. Post findings as a comment on this issue.