431 kB

Ctrl+K

4 contributors

History: 16 commits

Codeseys

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

84740d4 14 days ago

diloco
Wave 20: ModalSpawnExecutor — finish the Modal-backed serverless executor 14 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 17 days ago
ingestion
Wave 21: close both Wave 20 debt items — chat-template alignment + structural is_error 14 days ago
recipes
feat(trainer): ADR-008 Dr.GRPO config + SDPO strict-alignment guard 14 days ago
replaysim
Wave 14: close every Wave 13 review finding + 4 documentation files; Wave 14b: real PRIME-RL parity + multi-process DiLoCo convergence 17 days ago
tests
feat(hints): ADR-009 layered HintGenerator; accepted 14 days ago
trainer
feat(trainer): ADR-008 Dr.GRPO config + SDPO strict-alignment guard 14 days ago
README.md

1.98 kB
Wave 10 — packaging: composer_replication is now pip-installable 17 days ago
__init__.py

4.64 kB
Wave 12: close V1-V8 brief — GPU smoke, SDPO firing, real-trace e2e 17 days ago
batch.py

8.82 kB
Wave 12: close V1-V8 brief — GPU smoke, SDPO firing, real-trace e2e 17 days ago
hint_generator.py

12.1 kB
feat(hints): ADR-009 layered HintGenerator; accepted 14 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 17 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 17 days ago
teacher_replay.py

9.83 kB
Wave 10 — packaging: composer_replication is now pip-installable 17 days ago