File size: 4,500 Bytes
d124aee
 
 
 
 
 
 
043ea5c
572201a
 
 
3402390
 
d124aee
 
 
 
 
 
 
 
 
572201a
 
287b917
d124aee
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
418bed4
287b917
 
d124aee
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
572201a
287b917
572201a
287b917
 
3402390
a0384f7
d124aee
 
3402390
 
 
 
 
d124aee
da5ada1
a0384f7
d124aee
 
da5ada1
a0384f7
d124aee
 
da5ada1
 
3402390
a0384f7
da5ada1
3402390
da5ada1
d124aee
 
 
 
 
 
 
 
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
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
import gradio as gr
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
from huggingface_hub import ModelCard, DatasetCard, model_info, dataset_info
import logging
from typing import Tuple, Literal
import functools
import spaces
from cachetools import TTLCache
from cachetools.func import ttl_cache
import time
import os
os.environ['HF_TRANSFER'] = "1"
# Set up logging
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)

# Global variables
MODEL_NAME = "davanstrien/Smol-Hub-tldr"
model = None
tokenizer = None
device = None
CACHE_TTL = 6 * 60 * 60  # 6 hours in seconds
CACHE_MAXSIZE = 100

def load_model():
    global model, tokenizer, device
    logger.info("Loading model and tokenizer...")
    try:
        device = "cuda" if torch.cuda.is_available() else "cpu"
        tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME, use_fast=True)
        model = AutoModelForCausalLM.from_pretrained(MODEL_NAME)
        model = model.to(device)
        model.eval()
        return True
    except Exception as e:
        logger.error(f"Failed to load model: {e}")
        return False

@functools.lru_cache(maxsize=100)
def get_card_info(hub_id: str) -> Tuple[str, str]:
    """Get card information from a Hugging Face hub_id."""
    try:
        info = model_info(hub_id)
        card = ModelCard.load(hub_id)
        return "model", card.text
    except Exception as e:
        logger.error(f"Error fetching model card for {hub_id}: {e}")
        try:
            info = dataset_info(hub_id)
            card = DatasetCard.load(hub_id)
            return "dataset", card.text
        except Exception as e:
            logger.error(f"Error fetching dataset card for {hub_id}: {e}")
            raise ValueError(f"Could not find model or dataset with id {hub_id}")

@spaces.GPU
def _generate_summary_gpu(card_text: str, card_type: str) -> str:
    """Internal function that runs on GPU."""
    # Determine prefix based on card type
    prefix = "<MODEL_CARD>" if card_type == "model" else "<DATASET_CARD>"

    # Format input according to the chat template
    messages = [{"role": "user", "content": f"{prefix}{card_text}"}]
    inputs = tokenizer.apply_chat_template(
        messages, add_generation_prompt=True, return_tensors="pt"
    )
    inputs = inputs.to(device)

    # Generate with optimized settings
    with torch.no_grad():
        outputs = model.generate(
            inputs,
            max_new_tokens=60,
            pad_token_id=tokenizer.pad_token_id,
            eos_token_id=tokenizer.eos_token_id,
            temperature=0.4,
            do_sample=True,
            use_cache=True,
        )

    # Extract and clean up the summary
    input_length = inputs.shape[1]
    response = tokenizer.decode(outputs[0][input_length:], skip_special_tokens=False)

    # Extract just the summary part
    try:
        summary = response.split("<CARD_SUMMARY>")[-1].split("</CARD_SUMMARY>")[0].strip()
    except IndexError:
        summary = response.strip()

    return summary

@ttl_cache(maxsize=CACHE_MAXSIZE, ttl=CACHE_TTL)
def generate_summary(card_text: str, card_type: str) -> str:
    """Cached wrapper for generate_summary with TTL."""
    return _generate_summary_gpu(card_text, card_type)

def summarize(hub_id: str = "") -> str:
    """Interface function for Gradio. Returns JSON format."""
    try:
        if hub_id:
            # Fetch and infer card type automatically
            card_type, card_text = get_card_info(hub_id)
            # Use the cached wrapper
            summary = generate_summary(card_text, card_type)
            return f'{{"summary": "{summary}", "type": "{card_type}", "hub_id": "{hub_id}"}}'
        else:
            error_msg = "Error: Hub ID must be provided"
            return f'{{"error": "{error_msg}"}}'

    except Exception as e:
        error_msg = str(e)
        return f'{{"error": "{error_msg}"}}'

def create_interface():
    interface = gr.Interface(
        fn=summarize,
        inputs=gr.Textbox(label="Hub ID", placeholder="e.g., huggingface/llama-7b"),
        outputs=gr.JSON(label="Output"),
        title="Hugging Face Hub TLDR Generator",
        description="Generate concise summaries of model and dataset cards from the Hugging Face Hub.",
    )
    return interface

if __name__ == "__main__":
    if load_model():
        interface = create_interface()
        interface.launch()
    else:
        print("Failed to load model. Please check the logs for details.")