rhyliieee/LLAMA3-MED-v1.2

Finetuned a pretrained Model with Lora, resize the base model's embeddings, then load Peft Model with the resized base model.

"""

add special tokens to the tokenizer and base model before merging peft with base

open_tokenizer.add_special_tokens({ "additional_special_tokens": ["<|start_header_id|>", "<|end_header_id|>", "<|eot_id|>"] }) base_model.resize_token_embeddings(len(open_tokenizer))

reload the peft model with resized token embedding of base model

peft_model = PeftModel.from_pretrained(base_model, "rhyliieee/LLaMA3-8Bit-Lora-Med-v1",)

perform merging

merged_peft_base_with_special_tokens = peft_model.merge_and_unload() """