578 kB

Ctrl+K

4 contributors

History: 23 commits

Codeseys

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

aae66fa 12 days ago

datagen
fix(phase-8): close all 5 cross-family final-verify findings + regression tests 12 days ago
diloco
Wave 20: ModalSpawnExecutor — finish the Modal-backed serverless executor 13 days ago
distillation
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 15 days ago
ingestion
Wave 21: close both Wave 20 debt items — chat-template alignment + structural is_error 13 days ago
integrations
fix(phase-8): close all 5 cross-family final-verify findings + regression tests 12 days ago
recipes
feat(trainer): ADR-008 Dr.GRPO config + SDPO strict-alignment guard 12 days ago
replaysim
Wave 14: close every Wave 13 review finding + 4 documentation files; Wave 14b: real PRIME-RL parity + multi-process DiLoCo convergence 16 days ago
tests
feat(wave-a): close ADR-011 (SDPO alignment indices) + ADR-012 (review findings) 12 days ago
trainer
feat(trainer): policy-optimization objective MENU (ADR-014) 12 days ago
README.md

1.98 kB
Wave 10 — packaging: composer_replication is now pip-installable 16 days ago
__init__.py

4.64 kB
Wave 12: close V1-V8 brief — GPU smoke, SDPO firing, real-trace e2e 16 days ago
batch.py

8.82 kB
Wave 12: close V1-V8 brief — GPU smoke, SDPO firing, real-trace e2e 16 days ago
hint_generator.py

16.8 kB
feat(wave-a): close ADR-011 (SDPO alignment indices) + ADR-012 (review findings) 12 days ago
loss.py

13.4 kB
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 15 days ago
opsd.py

7.66 kB
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 15 days ago
teacher_replay.py

9.83 kB
Wave 10 — packaging: composer_replication is now pip-installable 16 days ago