Views
No views yet
1!pip install -q -U trl transformers accelerate git+https://github.com/huggingface/peft.git
2!pip install -q datasets bitsandbytes einops wandb sentencepiece transformers_stream_generator tiktoken
3
4from transformers import AutoModelForCausalLM, AutoTokenizer
5import torch
6
7tokenizer = AutoTokenizer.from_pretrained("TinyPixel/qwen-1.8B-guanaco", trust_remote_code=True)
8model = AutoModelForCausalLM.from_pretrained("TinyPixel/qwen-1.8B-guanaco", torch_dtype=torch.bfloat16, device_map="auto", trust_remote_code=True)
9
10device = "cuda:0"
11
12from transformers import StoppingCriteria, StoppingCriteriaList
13
14stop_token_ids = [[14374, 11097, 25], [14374, 21388, 25]]
15stop_token_ids = [torch.LongTensor(x).to(device) for x in stop_token_ids]
16
17from transformers import StoppingCriteria, StoppingCriteriaList
18
19class StopOnTokens(StoppingCriteria):
20 def __call__(self, input_ids: torch.LongTensor, scores: torch.FloatTensor, **kwargs) -> bool:
21 for stop_ids in stop_token_ids:
22 if torch.eq(input_ids[0][-len(stop_ids):], stop_ids).all():
23 return True
24 return False
25
26stopping_criteria = StoppingCriteriaList([StopOnTokens()])
27
28text = '''### Human: what is the difference between a dog and a cat on a biological level?
29### Assistant:'''
30
31inputs = tokenizer(text, return_tensors="pt").to(device)
32outputs = model.generate(**inputs,
33 max_new_tokens=512,
34 stopping_criteria=stopping_criteria,
35 do_sample=True,
36 top_p=0.95,
37 temperature=0.7,
38 top_k=50)
39
40print(tokenizer.decode(outputs[0], skip_special_tokens=False)