Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,161 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
import torch
|
3 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer
|
4 |
+
from huggingface_hub import ModelCard, DatasetCard, model_info, dataset_info
|
5 |
+
import logging
|
6 |
+
from typing import Tuple, Literal
|
7 |
+
import functools
|
8 |
+
|
9 |
+
# Set up logging
|
10 |
+
logging.basicConfig(level=logging.INFO)
|
11 |
+
logger = logging.getLogger(__name__)
|
12 |
+
|
13 |
+
# Global variables
|
14 |
+
MODEL_NAME = "davanstrien/Smol-Hub-tldr"
|
15 |
+
model = None
|
16 |
+
tokenizer = None
|
17 |
+
device = None
|
18 |
+
|
19 |
+
def load_model():
|
20 |
+
global model, tokenizer, device
|
21 |
+
logger.info("Loading model and tokenizer...")
|
22 |
+
try:
|
23 |
+
device = "cuda" if torch.cuda.is_available() else "cpu"
|
24 |
+
tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME, use_fast=True)
|
25 |
+
model = AutoModelForCausalLM.from_pretrained(MODEL_NAME)
|
26 |
+
model = model.to(device)
|
27 |
+
model.eval()
|
28 |
+
return True
|
29 |
+
except Exception as e:
|
30 |
+
logger.error(f"Failed to load model: {e}")
|
31 |
+
return False
|
32 |
+
|
33 |
+
@functools.lru_cache(maxsize=100)
|
34 |
+
def get_card_info(hub_id: str) -> Tuple[str, str]:
|
35 |
+
"""Get card information from a Hugging Face hub_id."""
|
36 |
+
try:
|
37 |
+
info = model_info(hub_id)
|
38 |
+
card = ModelCard.load(hub_id)
|
39 |
+
return "model", card.text
|
40 |
+
except Exception as e:
|
41 |
+
logger.error(f"Error fetching model card for {hub_id}: {e}")
|
42 |
+
try:
|
43 |
+
info = dataset_info(hub_id)
|
44 |
+
card = DatasetCard.load(hub_id)
|
45 |
+
return "dataset", card.text
|
46 |
+
except Exception as e:
|
47 |
+
logger.error(f"Error fetching dataset card for {hub_id}: {e}")
|
48 |
+
raise ValueError(f"Could not find model or dataset with id {hub_id}")
|
49 |
+
|
50 |
+
@functools.lru_cache(maxsize=100)
|
51 |
+
def generate_summary(card_text: str, card_type: str) -> str:
|
52 |
+
"""Generate a summary for the given card text."""
|
53 |
+
# Determine prefix based on card type
|
54 |
+
prefix = "<MODEL_CARD>" if card_type == "model" else "<DATASET_CARD>"
|
55 |
+
|
56 |
+
# Format input according to the chat template
|
57 |
+
messages = [{"role": "user", "content": f"{prefix}{card_text}"}]
|
58 |
+
inputs = tokenizer.apply_chat_template(
|
59 |
+
messages, add_generation_prompt=True, return_tensors="pt"
|
60 |
+
)
|
61 |
+
inputs = inputs.to(device)
|
62 |
+
|
63 |
+
# Generate with optimized settings
|
64 |
+
with torch.no_grad():
|
65 |
+
outputs = model.generate(
|
66 |
+
inputs,
|
67 |
+
max_new_tokens=60,
|
68 |
+
pad_token_id=tokenizer.pad_token_id,
|
69 |
+
eos_token_id=tokenizer.eos_token_id,
|
70 |
+
temperature=0.4,
|
71 |
+
do_sample=True,
|
72 |
+
use_cache=True,
|
73 |
+
)
|
74 |
+
|
75 |
+
# Extract and clean up the summary
|
76 |
+
input_length = inputs.shape[1]
|
77 |
+
response = tokenizer.decode(outputs[0][input_length:], skip_special_tokens=False)
|
78 |
+
|
79 |
+
# Extract just the summary part
|
80 |
+
try:
|
81 |
+
summary = response.split("<CARD_SUMMARY>")[-1].split("</CARD_SUMMARY>")[0].strip()
|
82 |
+
except IndexError:
|
83 |
+
summary = response.strip()
|
84 |
+
|
85 |
+
return summary
|
86 |
+
|
87 |
+
def summarize(hub_id: str = "", card_type: str = "model", content: str = "") -> str:
|
88 |
+
"""Interface function for Gradio."""
|
89 |
+
try:
|
90 |
+
if hub_id:
|
91 |
+
# Fetch and validate card type
|
92 |
+
inferred_type, card_text = get_card_info(hub_id)
|
93 |
+
if card_type and card_type != inferred_type:
|
94 |
+
return f"Error: Provided card_type '{card_type}' doesn't match inferred type '{inferred_type}'"
|
95 |
+
card_type = inferred_type
|
96 |
+
elif content:
|
97 |
+
if not card_type:
|
98 |
+
return "Error: card_type must be provided when using direct content"
|
99 |
+
card_text = content
|
100 |
+
else:
|
101 |
+
return "Error: Either hub_id or content must be provided"
|
102 |
+
|
103 |
+
summary = generate_summary(card_text, card_type)
|
104 |
+
return summary
|
105 |
+
|
106 |
+
except Exception as e:
|
107 |
+
return f"Error: {str(e)}"
|
108 |
+
|
109 |
+
# Create the Gradio interface
|
110 |
+
def create_interface():
|
111 |
+
with gr.Blocks(title="Hub TLDR") as interface:
|
112 |
+
gr.Markdown("# Hugging Face Hub TLDR Generator")
|
113 |
+
gr.Markdown("Generate concise summaries of model and dataset cards from the Hugging Face Hub.")
|
114 |
+
|
115 |
+
with gr.Tab("Summarize by Hub ID"):
|
116 |
+
hub_id_input = gr.Textbox(
|
117 |
+
label="Hub ID",
|
118 |
+
placeholder="e.g., huggingface/llama-7b"
|
119 |
+
)
|
120 |
+
hub_id_type = gr.Radio(
|
121 |
+
choices=["model", "dataset"],
|
122 |
+
label="Card Type (optional)",
|
123 |
+
value="model"
|
124 |
+
)
|
125 |
+
hub_id_button = gr.Button("Generate Summary")
|
126 |
+
hub_id_output = gr.Textbox(label="Summary")
|
127 |
+
|
128 |
+
hub_id_button.click(
|
129 |
+
fn=summarize,
|
130 |
+
inputs=[hub_id_input, hub_id_type],
|
131 |
+
outputs=hub_id_output
|
132 |
+
)
|
133 |
+
|
134 |
+
with gr.Tab("Summarize Custom Content"):
|
135 |
+
content_input = gr.Textbox(
|
136 |
+
label="Content",
|
137 |
+
placeholder="Paste your model or dataset card content here...",
|
138 |
+
lines=10
|
139 |
+
)
|
140 |
+
content_type = gr.Radio(
|
141 |
+
choices=["model", "dataset"],
|
142 |
+
label="Card Type",
|
143 |
+
value="model"
|
144 |
+
)
|
145 |
+
content_button = gr.Button("Generate Summary")
|
146 |
+
content_output = gr.Textbox(label="Summary")
|
147 |
+
|
148 |
+
content_button.click(
|
149 |
+
fn=lambda content, card_type: summarize(content=content, card_type=card_type),
|
150 |
+
inputs=[content_input, content_type],
|
151 |
+
outputs=content_output
|
152 |
+
)
|
153 |
+
|
154 |
+
return interface
|
155 |
+
|
156 |
+
if __name__ == "__main__":
|
157 |
+
if load_model():
|
158 |
+
interface = create_interface()
|
159 |
+
interface.launch()
|
160 |
+
else:
|
161 |
+
print("Failed to load model. Please check the logs for details.")
|