gahow/aituner
0
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Distinguish proxy and exact state verdicts

Browse Source
This commit is contained in:
Gahow Wang
2026-07-19 19:12:51 +08:00
parent 3be6cd04ad
commit 27b2daf7ce
1 changed files with 10 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
runs/frontier-fidelity-envelope-v1/analyze_qwen235_fixed_pd_state.py
View File

@@ -695,10 +695,16 @@ def analyze(
if decision_contrast is None: if decision_contrast is None:
verdict = "STOP: real Running proxy has insufficient exact simulator support" verdict = "STOP: real Running proxy has insufficient exact simulator support"
elif decision_contrast["total"] < 0: elif decision_contrast["total"] < 0:
verdict = ( if exact_contrast is not None:
"State-composition mismatch is insufficient: after reweighting Frontier " verdict = (
"to measured real decode composition, it still predicts TP8 faster." "State-composition mismatch is insufficient: after reweighting Frontier "
) "to measured real decode composition, it still predicts TP8 faster."
)
else:
verdict = (
"Active-batch-count mismatch is insufficient: after reweighting Frontier "
"to the frozen-real Running proxy, it still predicts TP8 faster."
)
else: else:
verdict = ( verdict = (
"Coarse active-batch state can flip Frontier's ordering, but iteration-level " "Coarse active-batch state can flip Frontier's ordering, but iteration-level "