Codeseys
/

composer-replication-framework

Reinforcement Learning

Model card Files Files and versions

composer-replication-framework / docs /adrs

Commit History

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

9a2ce20

Baladithya Balamurugan Claude Fable 5 commited on 2 days ago

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 3 days ago

docs(wave3): add OVERVIEW.md, index ADR-014, fold in adversarial-review fixes

e130879

Codeseys Claude Opus 4.8 (1M context) commited on 3 days ago

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

aae66fa

Codeseys commited on 13 days ago

docs(adr-013): close acceptance-gate boxes 1-5; only user-gated spend remains

7b34ebf

Codeseys commited on 14 days ago

feat(b4-gpu+b6): GPU train-proof on A10G + docker-gated substrate E2E test

c009712

Codeseys commited on 14 days ago

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

d02d724

Codeseys commited on 14 days ago

architect: ADR-011/012/013 + research (alignment-index fix, review-findings closure, LMA channel-ladder)

b4a584a

Codeseys commited on 14 days ago

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

185cce2

Codeseys commited on 14 days ago

feat(datagen): ADR-010 FeatureDeletionEnv synthetic-data subsystem; accepted

9336af3

Codeseys commited on 14 days ago

feat(hints): ADR-009 layered HintGenerator; accepted

84740d4

Codeseys commited on 14 days ago

feat(trainer): ADR-008 gate-3 live GRPO+SDPO smoke PASS; ADR-008 accepted

2a34df4

Codeseys commited on 14 days ago

docs(adr): add ADR-008/009/010 (Dr.GRPO+SDPO, layered hints, FeatureDeletionEnv)

36ab61e

Codeseys commited on 14 days ago

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 17 days ago

Wave 14: close every Wave 13 review finding + 4 documentation files; Wave 14b: real PRIME-RL parity + multi-process DiLoCo convergence

d9dd3a5

Codeseys commited on 17 days ago

Wave 13: serverless DiLoCo + replaysim normalization + 3 distillation losses + PRIME-RL + Monarch

b266c31

Codeseys commited on 17 days ago

Wave 11: cross-model adversarial review + honest down-revision

f16fa23

Codeseys commited on 17 days ago

Wave 7: Phase 2-4 of deep work loop — backlog, parallel research, three ADRs

ac4bfb4

Codeseys commited on 17 days ago