Inference Example:
from peft import PeftModel, PeftConfig
from transformers import AutoModelForCausalLM, AutoTokenizer
peft_model_id = "edu-linguistic/opt-1.3b-edu-sft"
model_name = 'facebook/opt-1.3b'
config = PeftConfig.from_pretrained(peft_model_id)
model = AutoModelForCausalLM.from_pretrained(model_name)
model = PeftModel.from_pretrained(model, peft_model_id)
tokenizer = AutoTokenizer.from_pretrained(model_name)
question = "<|prompter|> Consider the following function: f(x1, x2) = ln(x1). This function is…"
question = tokenizer.encode(question, return_tensors='pt')
generation_kwargs = {
"do_sample": True,
"top_k": 0,
"top_p": 0.9,
"bos_token_id": tokenizer.bos_token_id,
"pad_token_id": tokenizer.pad_token_id,
"eos_token_id": tokenizer.eos_token_id,
"num_return_sequences": 1,
"min_new_tokens": 10,
"max_new_tokens": 512,
}
response = model.generate(input_ids=question, **generation_kwargs)
response = tokenizer.decode(response[0],
skip_special_tokens=False,
clean_up_tokenization_spaces=False
)
print(response)