File size: 2,669 Bytes
1b3b9d7
 
9a26d37
1b3b9d7
 
 
 
 
ed96116
d73e65e
 
1b3b9d7
 
 
 
 
 
 
 
 
 
 
 
 
ec93e16
1b3b9d7
 
2fc1738
 
 
 
 
 
 
0ca8c6e
1b3b9d7
 
 
 
 
 
 
 
d73e65e
 
1b3b9d7
 
 
 
 
 
 
 
 
 
 
 
 
 
 
d73e65e
2fc1738
 
1b3b9d7
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
import gradio as gr
import torch
import spaces
from transformers import AutoModelForCausalLM, AutoTokenizer
from transformers import StoppingCriteria, StoppingCriteriaList, TextIteratorStreamer
from threading import Thread

# Loading the tokenizer and model from Hugging Face's model hub.
if torch.cuda.is_available():
    tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen1.5-7B-Chat")
    model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen1.5-7B-Chat", torch_dtype=torch.float16, device_map="auto")


# Defining a custom stopping criteria class for the model's text generation.
class StopOnTokens(StoppingCriteria):
    def __call__(self, input_ids: torch.LongTensor, scores: torch.FloatTensor, **kwargs) -> bool:
        stop_ids = [2]  # IDs of tokens where the generation should stop.
        for stop_id in stop_ids:
            if input_ids[0][-1] == stop_id:  # Checking if the last generated token is a stop token.
                return True
        return False


# Function to generate model predictions.
@spaces.GPU(duration=600)
def predict(message, history):
    stop = StopOnTokens()
    conversation = []
    
    for user, assistant in history:
        conversation.extend([{"role": "user", "content": user}, {"role": "assistant", "content": assistant}])
        
    conversation.append({"role": "user", "content": message})
    prompt = tokenizer.apply_chat_template(conversation, tokenize=False, add_generation_prompt=True)
    model_inputs = tokenizer(prompt, return_tensors="pt").to("cuda")
    streamer = TextIteratorStreamer(tokenizer, timeout=10., skip_prompt=True, skip_special_tokens=True)
    generate_kwargs = dict(
        model_inputs,
        streamer=streamer,
        max_new_tokens=1024,
        do_sample=True,
        top_p=0.95,
        top_k=50,
        temperature=0.7,
        repetition_penalty=1.0,
        num_beams=1,
        stopping_criteria=StoppingCriteriaList([stop])
    )
    t = Thread(target=model.generate, kwargs=generate_kwargs)
    t.start()  # Starting the generation in a separate thread.
    partial_message = ""
    for new_token in streamer:
        partial_message += new_token
        if '</s>' in partial_message:  # Breaking the loop if the stop token is generated.
            break
        yield partial_message


# Setting up the Gradio chat interface.
gr.ChatInterface(predict,
                 title="Qwen1.5 7B Chat Demo",
                 description="Warning. All answers are generated and may contain inaccurate information.",
                 examples=['How do you cook fish?', 'Who is the president of the United States?']
                 ).launch()  # Launching the web interface.