pullrequests Search Results · language:Dune language:Python language:Python language:Java language:JavaScript language:Java
Filter by
182M results
Summary
- finalize taskset hook — the symmetric counterpart to setup, run after the harness finishes and before scoring, while
the runtime is still live. Override it for per-rollout runtime work ...
Bumps the actions group with 5 updates in the / directory:
| Package | From | To |
| --- | --- | --- |
| actions/checkout | 6.0.2 | 6.0.3 |
| actions/upload-artifact | 7.0.0 | 7.0.1 |
| dawidd6/action-download-artifact ...
dependencies
github_actions
PR1 — Live-progress record + richer busy-ack
Changes
app/main.py:
- Added _live_progress: dict[session_id, dict] — per-session record updated by _run_tool_round_loop
- Added _render_progress(session_id) ...
Adds the per-cell runtime params for SPEED-bench gemma-4-E4B-it / MTP / vLLM cell t0_d7.
Sweep: gemma-4-E4B-it_mtp_vllm_t0_d7
Résumé
Ajoute un Makefile à la racine comme point d entrée unique pour le développement.
- make dev — lance backend (tsx watch, :8081) et frontend (Vite, :3000) ensemble ; Ctrl+C arrête les deux. Vérifie ...
Summary
- _write_live_eval_run: after INSERT to ds_eval_runs, UPDATE eval_registry fields last_run_at, last_run_id,
rubric_score (stored as 0–100 integer) so ds eval registry show reflects live-run ...
Pending decision - CI checking in advance Reverts vllm-project/tpu-inference#2711
ready
Follow-up to #2 (post-merge work rebased onto main, preserving the explicit-\n semibold-lead refinement from the
squash).
Library
- save_chart(__file__) — the example save epilogue as one call (was ...