MoE LoRA: auto-target per-expert Linear experts (gpt-oss 4bit) instead of leaving them frozen - #6936

Merged

danielhanchen merged 3 commits into

mainfrom

fix/moe-target-honest-missing-experts

Jul 8, 2026

Commits on Jul 8, 2026

models: auto-target per-expert Linear MoE experts for LoRA (gpt-oss 4bit)
danielhanchen
committed
2129fc7
models: scope per-expert MoE targets, keep repeat get_peft_model idempotent, warn on old zoo
danielhanchen
committed
5cf6e3c
[pre-commit.ci] auto fixes from pre-commit.com hooks
pre-commit-ci[bot]
committed
8a93a5d