metadata
license: apache-2.0
language:
- en
base_model:
- meta-llama/Llama-3.1-8B-Instruct
- meta-llama/Llama-2-7b
pipeline_tag: question-answering
tags:
- medical
- biology
- genetics
- bioinformatics
GP-GTP is an open-weight genetic-phenotype knowledge language model. For "medical-genetic-information".
Arvix version: arXiv:2409.09825
Usage
from transformers import AutoModelForCausalLM, BitsAndBytesConfig, HfArgumentParser, TrainingArguments
from peft import AutoPeftModelForCausalLM
from peft import PeftModel
from peft import LoraConfig, get_peft_model
#init
parser = HfArgumentParser(ScriptArguments)
script_args = parser.parse_args_into_dataclasses()[0]
# specific the model to load
# For GP-GPT small:
script_args.model_name = "meta-llama/Llama-2-7b"
script_args.peft_model_id = "./small/"
# For GP-GPT base:
script_args.model_name = "meta-llama/Meta-Llama-3.1-8B"
script_args.peft_model_id = "./base/"
# Cache model
model = AutoModelForCausalLM.from_pretrained(
script_args.model_name,
#quantization_config=quantization_config, # activate when using quantization setting
device_map=device_map,
torch_dtype=torch_dtype,
use_auth_token=False,
)
#load PEFT adapter
if script_args.peft_model_id is not None:
peft_model_id = script_args.peft_model_id
model = PeftModel.from_pretrained(model, peft_model_id)
model = model.merge_and_unload()