Carry reward and per-token model_version, log reward/version metrics - #562

Merged

jlamypoirier merged 10 commits into

mainfrom

jlp_reward_model_version

Jul 10, 2026

Commits on Jul 8, 2026

Carry reward and per-token model_version, log reward/version metrics

jlamypoirier
and
claude
committed
5975c8d

Commits on Jul 10, 2026

Merge remote-tracking branch 'origin/main' into jlp_reward_model_version
jlamypoirier
committed
e534dca
Merge remote-tracking branch 'origin/main' into jlp_reward_model_version
jlamypoirier
committed
829d676
Review fixes: inline single-use target helper, hoist inf sentinels, trim comments

jlamypoirier
and
claude
committed
e08726e
Log reward metric as train_samples_reward to flag filtering bias

jlamypoirier
and
claude
committed
28b3fd6
Log staleness (documents_seen - model_version) instead of raw model version

jlamypoirier
and
claude
committed
b225da1
Review fixes: rename masked->document_weight, loop var key->data_key

jlamypoirier
and
claude
committed
9daf611
Register reward/staleness metrics unconditionally (fields are required)

jlamypoirier
and
claude
committed
f64ce7c
Review fixes: int-domain staleness subtraction, naming/comment nits

jlamypoirier
and
claude
committed
91f674d
Merge branch 'main' into jlp_reward_model_version
jlamypoirier
authored
6d69ff5