File size: 1,612 Bytes
e2fac8d
 
6293678
29e0785
6293678
e2fac8d
 
 
 
 
 
29e0785
e2fac8d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
29e0785
e2fac8d
 
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
import torch
from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig
import gradio as gr
import spaces

model_id = "meta-llama/Llama-Guard-3-8B-INT8"
device = "cuda" if torch.cuda.is_available() else "cpu"
dtype = torch.bfloat16

quantization_config = BitsAndBytesConfig(load_in_8bit=True)

@spaces.GPU
def load_model():
    tokenizer = AutoTokenizer.from_pretrained(model_id)
    model = AutoModelForCausalLM.from_pretrained(
        model_id, 
        torch_dtype=dtype, 
        device_map=device, 
        quantization_config=quantization_config
    )
    return tokenizer, model

tokenizer, model = load_model()

def moderate(user_input, assistant_response):
    chat = [
        {"role": "user", "content": user_input},
        {"role": "assistant", "content": assistant_response},
    ]
    input_ids = tokenizer.apply_chat_template(chat, return_tensors="pt").to(device)
    output = model.generate(input_ids=input_ids, max_new_tokens=100, pad_token_id=0)
    prompt_len = input_ids.shape[-1]
    return tokenizer.decode(output[0][prompt_len:], skip_special_tokens=True)

def gradio_moderate(user_input, assistant_response):
    return moderate(user_input, assistant_response)

iface = gr.Interface(
    fn=gradio_moderate,
    inputs=[
        gr.Textbox(lines=3, label="User Input"),
        gr.Textbox(lines=3, label="Assistant Response")
    ],
    outputs=gr.Textbox(label="Moderation Result"),
    title="Llama Guard Moderation",
    description="Enter a user input and an assistant response to check for content moderation."
)

if __name__ == "__main__":
    iface.launch()