jekunz/smollm-135m-lora-fineweb-faroese
Adapter vs HuggingFaceTB/SmolLM2-135M-Instruct. jekunz/smollm-135m-lora-fineweb-faroese is a LoRA adapter, not full weights: rank 256, on down_proj, gate_proj, o_proj, up_proj in 30 layers (120 adapted matrices). It modifies HuggingFaceTB/SmolLM2-135M-Instruct at load time.