File size: 638 Bytes
41f9ecd
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
generation_config = GenerationConfig(
    temperature=0.1,
    top_p=0.75,
    num_beams=4,
)

def evaluate(instruction, input=None):
    prompt = generate_prompt(instruction, input)
    inputs = tokenizer(prompt, return_tensors="pt")
    input_ids = inputs["input_ids"].cuda()
    generation_output = model.generate(
        input_ids=input_ids,
        generation_config=generation_config,
        return_dict_in_generate=True,
        output_scores=True,
        max_new_tokens=256
    )
    for s in generation_output.sequences:
        output = tokenizer.decode(s)
        print("Response:", output.split("### Response:")[1].strip())