Spaces:
Running
Running
Commit History
2 days ago
-
refactor: centralize all settings in app.py
- add a SETTINGS block at the top of app.py for model count, concurrency, max tokens, request timeout, results directory, discovery paging, and auto-run - add smoke.configure() so app.py applies those values at import time - remove environment-variable configuration for model limit, concurrency, tokens, timeout, paging, and results root - make fetch_trending_models/test_model/run_smoke resolve configured defaults at call time instead of binding them at import - replace the UI model-count slider with a read-only settings panel - add AUTO_RUN_ON_START, which runs the test on page load when a token exists - keep the Run button and route it through the configured values - remove now-unused imports from app.py - document the SETTINGS block in the README57dc987 -
feat: add inference smoke runner
- add test_model() which tries each candidate with :cheapest first - retry once with :fastest when the first attempt fails or returns no choices - treat only HTTP 200 with non-empty choices as True (pass) - classify HTTP status, timeout, and provider errors per attempt - redact the bearer token from every stored error message - use one InferenceClient per worker thread to avoid sharing a session - run models concurrently (default 3) with bounded 8-token chat requests - persist each run to /data/hf-inference-smoke/runs/<run_id>.json and latest.json - add run listing and loading with run_id validation to block traversal - add list_runs()/load_run()/load_latest_run() helpers for the UI799874f -
feat: add trending model discovery
- add smoke.py with persistent-storage paths under /data/hf-inference-smoke - query the Hub API for conversational models with inference_provider=all - sort by trendingScore and scan beyond the requested limit - filter out models without at least one live provider - save discovery snapshots and latest.json atomically - add token validation via the whoami-v2 endpoint - pin gradio, huggingface_hub, and requests in requirements.txt - ignore local runtime and agent artifactsad0d42a