[fix]: respect auto_model param in FastLlamaModel.from_pretrained for embedding models - #6886

Closed

InfoSage05 wants to merge 5 commits into

unslothai:mainfrom

InfoSage05:fix/6881-embedding-auto-model

Commits on Jul 5, 2026

fix: respect auto_model param in FastLlamaModel.from_pretrained

Ayushman Paul
authored and
Ayushman Paul
committed
a26f621
fix: use PyTorch QK norm for embedding models to match stock HF precision

Ayushman Paul
authored and
Ayushman Paul
committed
169e603

Commits on Jul 7, 2026

fix: keep UNSLOTH_EMBEDDING_HIGH_PRECISION active + guard CausalLM-only paths

Ayushman Paul
authored and
Ayushman Paul
committed
e5a94f5
fix: guard None auto_model with fallback to AutoModelForCausalLM

Ayushman Paul
authored and
Ayushman Paul
committed
0426ee4
Merge branch 'main' into fix/6881-embedding-auto-model
InfoSage05
authored
ac21014