Codeseys
/

composer-replication-framework

Reinforcement Learning

Model card Files Files and versions

composer-replication-framework / examples

Ctrl+K

Ctrl+K

4 contributors

History: 14 commits

Codeseys's picture

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

c009712 about 1 month ago

altered_minds_channel_ladder
feat(b4-gpu+b6): GPU train-proof on A10G + docker-gated substrate E2E test about 1 month ago
composer_grpo_sdpo_smoke
feat(trainer): ADR-008 gate-3 live GRPO+SDPO smoke PASS; ADR-008 accepted about 1 month ago
gsm8k_grpo
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 about 1 month ago
gsm8k_grpo_with_sdpo
Wave 17: close all 5 audit FLAGs + SDPO context alignment + serverless re-exports about 1 month ago
qwen_05b_quickstart
Wave 12: close V1-V8 brief — GPU smoke, SDPO firing, real-trace e2e about 1 month ago
sdpo_real_trace_train_smoke
examples: add sdpo_real_trace_train_smoke — close the forward+backward+step link about 1 month ago
sdpo_with_real_traces
Wave 19: production-grade SDPO via ComposerDataCollator + adapter + collator fixes about 1 month ago
sdpo_with_real_traces_production
Wave 21: close both Wave 20 debt items — chat-template alignment + structural is_error about 1 month ago
validate_real_trace_alignment
Wave 21b: skip zero-signal SDPO on empty-recovery error turns + real-trace validation about 1 month ago
README.md

3.12 kB
Wave 19: production-grade SDPO via ComposerDataCollator + adapter + collator fixes about 1 month ago