fix(reflect): support continuous reward scores in failure filtering

Browse Source

not r.get("hard") treats non-zero floats as success.
Add explicit float threshold check (< 1e-9).
Backward compatible with binary hard=0/1.

...

This commit is contained in:

zq

2026-05-29 19:04:42 +08:00

parent afb552008b

commit a62ec857f1

1 changed files with 588 additions and 588 deletions

Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL

Download Patch File Download Diff File Expand all files Collapse all files

skillopt/gradient/reflect.py

+588 -588

File diff suppressed because it is too large Load Diff