Hugging Face's logo

Codeseys
/
composer-replication-framework

Reinforcement Learning
Transformers
English
post-training
distillation
agentic-coding
composer-2.5
cursor
kimi-k2
grpo
dapo
diloco
openenv
trl
verl
research
methodology
Model card Files
xet
 Community
composer-replication-framework / composer_replication /trainer

Commit History

Wave 21: Stage-0 dataset pipeline — swesmith engine, rollout harness, gates, contract
9a2ce20

Baladithya Balamurugan Claude Fable 5 commited on

Wave 20: Tier-0 fidelity fixes — k1-in-reward KL + Composer-2 behavior rewards
41289bf

Baladithya Balamurugan Claude Opus 4.8 (1M context) commited on

Wave 3: close the HIGH review findings (kill-switch wiring, HeldoutSplit, EKS entrypoint bug)
bd0c358

Baladithya Balamurugan Claude Opus 4.8 (1M context) commited on

Wave 1: fix 8 failing tests + unblock Docker E2E + dep/doc debt
c11cf49

Baladithya Balamurugan Claude Opus 4.8 (1M context) commited on

feat(trainer): policy-optimization objective MENU (ADR-014)
aae66fa

Codeseys commited on

fix(phase-8): close all 5 cross-family final-verify findings + regression tests
678d10b

Codeseys commited on

feat(wave-a): close ADR-011 (SDPO alignment indices) + ADR-012 (review findings)
d02d724

Codeseys commited on

review+fix: cross-family adversarial ADR review (owed item) + remediation
185cce2

Codeseys commited on

feat(trainer): ADR-008 Dr.GRPO config + SDPO strict-alignment guard
bde5c5e

Codeseys commited on

Wave 21b: skip zero-signal SDPO on empty-recovery error turns + real-trace validation
d61036a

Codeseys commited on

Wave 21: close both Wave 20 debt items — chat-template alignment + structural is_error
6806cf7

Codeseys commited on

Wave 19: production-grade SDPO via ComposerDataCollator + adapter + collator fixes
03bf323

Codeseys commited on

Wave 15: 4-angle multi-model self-critique caught 2 math BLOCKERs in primary loss kernels; fixed against upstream byte-for-byte + GSM8K example + ergonomics
e5add15

Codeseys commited on

Wave 10 — packaging: composer_replication is now pip-installable
ac05fbf

Codeseys commited on