Phase 2: ProTrain integration with Axolotl perf features (M0–M6C closed)#21

Closed

thad0ctor wants to merge 43 commits into

protrain-optim-checkpoint-phase2-mode-cfrom

protrain-phase2-integration

Commits on May 9, 2026

chore(protrain): fix pre-existing mypy on model_wrapper:386 + formatting

thad0ctor
and
claude
committed
feat(protrain): enable FlashAttention in canonical LoRA example (M4)

thad0ctor
and
claude
committed
feat(protrain): allow load_in_8bit / load_in_4bit (M2+M3 Mode A)

thad0ctor
and
claude
committed
feat(protrain): integrate fused LoRA kernels via container hooks (M1)

thad0ctor
and
claude
committed
feat(protrain): add bnb 8-bit AdamW optimizer adapter (M2.5)

thad0ctor
and
claude
committed
feat(protrain): reject unsupported optimizers at config load (M6B)

thad0ctor
and
claude
committed

Commits on May 10, 2026

Commits on May 11, 2026

Commits on May 12, 2026

Commits on May 13, 2026

fix(test): test_chunk_optim_shutdown caplog → mock.patch on LOG (CI flake)

thad0ctor
and
claude
committed

Commits on May 21, 2026