Views
No views yet
1import torch
2from transformers import AutoTokenizer, AutoModelForCausalLM
3
4device = "cuda"
5
6model = AutoModelForCausalLM.from_pretrained(
7 'Kendamarron/Tokara-0.5B-Chat-v0.1',
8 torch_dtype=torch.bfloat16,
9 device_map=device,
10)
11tokenizer = AutoTokenizer.from_pretrained('Kendamarron/Tokara-0.5B-Chat-dolly-jimba')
12
13messages = [
14 {"role": "system", "content": "あなたは誠実で優秀な日本人のアシスタントです。"},
15 {"role": "user", "content": "野菜は体にいいですか？"}
16]
17text = tokenizer.apply_chat_template(
18 messages,
19 tokenize=False,
20 add_generation_prompt=True
21)
22model_inputs = tokenizer([text], return_tensors="pt").to(device)
23generated_ids = model.generate(
24 model_inputs.input_ids,
25 max_new_tokens=256,
26 do_sample=True,
27 top_p=0.95,
28 top_k=40,
29 temperature=0.7,
30 repetition_penalty=1.1,
31 pad_token_id=tokenizer.eos_token_id,
32 eos_token_id=tokenizer.eos_token_id,
33 no_repeat_ngram_size=2
34 )
35generated_ids = [
36 output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
37]
38response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
39
40print(response)