File size: 4,219 Bytes
671d64d
e873032
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
671d64d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
e873032
671d64d
 
 
 
 
 
 
 
 
 
 
e873032
671d64d
 
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
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
import gradio as gr
from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer
import os
import os.path 
import pickle
import torch
from openai import OpenAI

base_model_id = "meta-llama/Llama-3.2-3B-Instruct"
model_id = "HiGenius/Headline-Generation-Model"

hf_token = os.environ.get('HF_TOKEN')
openai_api_key = os.environ.get('OPENAI_API_KEY')

device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')

def load_model():
    base_model = AutoModelForCausalLM.from_pretrained(base_model_id, use_auth_token=hf_token)
    model = PeftModel.from_pretrained(base_model, model_id, use_auth_token=hf_token).to(device)
    tokenizer = AutoTokenizer.from_pretrained(base_model_id, use_auth_token=hf_token)
    tokenizer.pad_token = tokenizer.eos_token
    tokenizer.padding_side='left'
    tokenizer.truncation_side="left"

    return tokenizer, model

def summarize_content(content):
    client = OpenAI(api_key=openai_api_key)
    response = client.chat.completions.create(
        model="gpt-4o",
        messages=[
            {"role": "system", "content": "Summarize the following article content concisely while preserving key information:"},
            {"role": "user", "content": content}
        ],
        max_tokens=600,
        temperature=0.3
    )
    return response.choices[0].message.content

tokenizer, model = load_model()

guideline_path = "./guidelines.txt"
with open(guideline_path, 'r', encoding='utf-8') as f:
    guidelines = f.read()

def process_prompt(tokenizer, content, video_summary = '', guidelines = None):
    # Check token lengths
    content_tokens = len(tokenizer.encode(content))
    total_tokens = content_tokens
    if video_summary:
        total_tokens += len(tokenizer.encode(video_summary))

    if content_tokens > 850 or total_tokens > 900:
        content = summarize_content(content)
    
    if guidelines:
        system_prompt = "You are a helpful assistant that writes engaging headlines. To maximize engagement, you may follow these proven guidelines:\n" + guidelines
    else:
        system_prompt = "You are a helpful assistant that writes engaging headlines."

    user_prompt = (
        f"Below is an article and its accompanying video summary:\n\n"
        f"Article Content:\n{content}\n\n"
        f"Video Summary:\n{'None' if video_summary == '' else video_summary}\n\n"
        f"Write ONLY a single engaging headline that accurately reflects the article. Do not include any additional text, explanations, or options."
    )
    messages = [
        {"role": "system", "content": system_prompt},
        {"role": "user", "content": user_prompt},
    ]
    prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
    return prompt

def generate_headlines(content, video_summary):
    if not content.strip():
        return "Please enter valid article content."
        
    if not video_summary.strip():
        video_summary = ''
        
    prompt = process_prompt(tokenizer, content, video_summary, guidelines)
    inputs = tokenizer(prompt, return_tensors="pt", truncation=True, max_length=1024).to(device)
    
    headlines = []
    for i in range(5):
        outputs = model.generate(**inputs,
                               max_new_tokens=60,
                               num_return_sequences=1,
                               do_sample=True,
                               temperature=0.7)
        response = tokenizer.decode(outputs[0][inputs['input_ids'].shape[1]:], skip_special_tokens=True)
        response = response.replace('"', '')
        headlines.append(f"Headline {i+1}: {response}")
    
    return "\n\n".join(headlines)

# Create Gradio interface
demo = gr.Interface(
    fn=generate_headlines,
    inputs=[
        gr.Textbox(label="Article Content", placeholder="Type the main content of the article here..."),
        gr.Textbox(label="Video Summary (Optional)", placeholder="Type the summary of the video related to the article...")
    ],
    outputs=gr.Textbox(label="Generated Headlines"),
    title="Article Headline Writer",
    description="Write catchy headlines from content and video summary."
)

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