[VIBE CODED - Review Only] E2E model competition support#442

Closed

msaroufim wants to merge 16 commits into

mainfrom

Commits on Feb 10, 2026

[VIBE CODED] Add e2e model competition support (vLLM fork benchmarking)
Mark Saroufim
committed
Address code review feedback
Mark Saroufim
committed
Add GitHub Actions support for model competitions
Mark Saroufim
committed
Fix test_backend.py: add score_ascending to expected config dicts
Mark Saroufim
committed
Add testing guide for model competitions (Modal-first E2E)
Mark Saroufim
committed

Commits on Feb 11, 2026

Use uv venv in model workflow and submission install
Mark Saroufim
committed
Add persistent model venv and fix setuptools-scm for tarballs
Mark Saroufim
committed
Simplify model workflow: local venv, no persistent paths
Mark Saroufim
committed
Fix vLLM server startup: conditional /models dir, capture logs
Mark Saroufim
committed
Pass HF_TOKEN to model workflow for gated model access
Mark Saroufim
committed
Update Llama-3.1-8B perplexity baseline to 1.80
Mark Saroufim
committed
Fix model competition E2E: CUDA 12.8 image, vllm bench CLI, overlay safety
Mark Saroufim
committed

Commits on Feb 12, 2026

Merge remote-tracking branch 'origin/main' into speedrun
Mark Saroufim
committed
Fix E2E bugs and add model competition testing guide
Mark Saroufim
committed
Add popcorn-cli submission instructions to model competition testing guide
Mark Saroufim
committed

Commits on Feb 13, 2026

Fix B200 model benchmark: cu128 torch, keep vLLM, use sys.executable
Mark Saroufim
committed