Files
probe_sitecustomize.py— put its directory onPYTHONPATHofvllm serveand setRACE_PROBE=log,geom RACE_PROBE_LOG=/path/probe.jsonl. Logs the runner'sreorder_batch_threshold, every builder's threshold, the rows of every Mamba2 prefill step ([request, num_computed_before, tokens_this_step]) and every batch swap. No behaviour change.RACE_PROBE=log,geom,reorderfixadditionally forces the runner threshold to1 + k(the monkeypatch used as the positive control).burst.py— load generator. `--mode pairs --pair-long-reps 1200 --pair-delay 1.5 --pair-delay-jitter 1.0