mirror of
https://github.com/wassname/evil_MoE.git
synced 2026-06-27 18:04:59 +08:00
probe_traj: side-by-side vanilla-vs-projected trajectory analyzer
Reads step files from both warmup-gen tags, prints per-step table broken into warmup-replay and student-gen phases, computes H1 delta on the gen-phase hack rate.
This commit is contained in:
@@ -212,6 +212,10 @@ probe-projected-replay steps="20":
|
||||
probe-uat:
|
||||
uv run python -m projected_grpo.probe_uat
|
||||
|
||||
# Trajectory comparator for the warmup-gen runs (vanilla vs projected).
|
||||
probe-traj:
|
||||
uv run python -m projected_grpo.probe_traj
|
||||
|
||||
# Phase 2 pilot analyzer: reads out/train_pilot_*.safetensors, prints trajectories
|
||||
# and per-arm aggregates, applies decision rules from spec2.md.
|
||||
phase2-analyze pattern="_pilot_*":
|
||||
|
||||
Reference in New Issue
Block a user