| from transformers import AutoModelForCausalLM, AutoTokenizer |
|
|
| model_name = "Qwen/Qwen2.5-0.5B-Instruct" |
|
|
| model = AutoModelForCausalLM.from_pretrained( |
| model_name, |
| torch_dtype="auto", |
| device_map="auto" |
| ) |
| tokenizer = AutoTokenizer.from_pretrained(model_name) |
| print("load model finished") |
| prompt = "Give me a short introduction to large language model." |
| messages = [ |
| {"role": "system", "content": "You are Qwen, created by Alibaba Cloud. You are a helpful assistant."}, |
| {"role": "user", "content": prompt} |
| ] |
| text = tokenizer.apply_chat_template( |
| messages, |
| tokenize=False, |
| add_generation_prompt=True |
| ) |
| print("after apply chat template:",text) |
| model_inputs = tokenizer([text], return_tensors="pt").to(model.device) |
|
|
| generated_ids = model.generate( |
| **model_inputs, |
| max_new_tokens=512 |
| ) |
| generated_ids = [ |
| output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids) |
| ] |
|
|
| response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0] |
