Views
No views yet
1from transformers import AutoModelForCausalLM, AutoTokenizer
2import torch
3device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
4
5MODEL_NAME = "DeepMount00/Llama-3.1-8b-Ita"
6
7model = AutoModelForCausalLM.from_pretrained(MODEL_NAME, torch_dtype=torch.bfloat16).eval()
8model.to(device)
9tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
10
11def generate_answer(prompt):
12 messages = [
13 {"role": "user", "content": prompt},
14 ]
15 model_inputs = tokenizer.apply_chat_template(messages, return_tensors="pt").to(device)
16 generated_ids = model.generate(model_inputs, max_new_tokens=200, do_sample=True,
17 temperature=0.001)
18 decoded = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)
19 return decoded[0]
20
21prompt = "Come si apre un file json in python?"
22answer = generate_answer(prompt)
23print(answer)| Metric | Value |
|---|---|
| Avg. | 28.23 |
| IFEval (0-Shot) | 79.17 |
| BBH (3-Shot) | 30.93 |
| MATH Lvl 5 (4-Shot) | 10.88 |
| GPQA (0-shot) | 5.03 |
| MuSR (0-shot) | 11.40 |
| MMLU-PRO (5-shot) | 31.96 |