Update app.py
Browse files
app.py
CHANGED
@@ -1,13 +1,17 @@
|
|
1 |
import gradio as gr
|
2 |
import torch
|
3 |
import json
|
4 |
-
from transformers import GPT2Config
|
5 |
from torch import nn
|
6 |
import requests
|
7 |
from pathlib import Path
|
|
|
|
|
|
|
|
|
|
|
8 |
|
9 |
class TextGenerator(nn.Module):
|
10 |
-
def __init__(self, vocab_size, embedding_dim, hidden_dim):
|
11 |
super().__init__()
|
12 |
self.embedding = nn.Embedding(vocab_size, embedding_dim)
|
13 |
self.lstm = nn.LSTM(embedding_dim, hidden_dim, batch_first=True)
|
@@ -19,19 +23,29 @@ class TextGenerator(nn.Module):
|
|
19 |
return self.fc(lstm_out)
|
20 |
|
21 |
def download_file(url, local_path):
|
22 |
-
|
23 |
-
|
|
|
24 |
Path(local_path).parent.mkdir(parents=True, exist_ok=True)
|
25 |
with open(local_path, 'wb') as f:
|
26 |
f.write(response.content)
|
27 |
-
|
28 |
-
|
|
|
|
|
29 |
|
30 |
def load_model_and_tokenizers():
|
31 |
# Create a local directory for downloaded files
|
32 |
cache_dir = Path("model_cache")
|
33 |
cache_dir.mkdir(exist_ok=True)
|
34 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
35 |
# URLs for the files
|
36 |
base_url = "https://huggingface.co/Omarrran/temp_data/raw/main"
|
37 |
files = {
|
@@ -45,56 +59,83 @@ def load_model_and_tokenizers():
|
|
45 |
for filename, url in files.items():
|
46 |
local_path = cache_dir / filename
|
47 |
if not local_path.exists():
|
48 |
-
|
49 |
download_file(url, local_path)
|
50 |
|
51 |
-
|
52 |
-
|
53 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
54 |
|
55 |
-
|
56 |
-
|
57 |
-
word_to_int
|
58 |
-
|
59 |
-
int_to_word
|
60 |
-
|
61 |
-
|
62 |
-
|
63 |
-
|
64 |
-
|
65 |
-
|
66 |
-
|
67 |
-
|
68 |
-
# Load model weights
|
69 |
-
model.load_state_dict(torch.load(cache_dir / "model.pt", map_location=torch.device('cpu')))
|
70 |
-
model.eval()
|
71 |
|
72 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
73 |
|
74 |
def generate_text(prompt, max_length=100):
|
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 |
# Create Gradio interface
|
100 |
iface = gr.Interface(
|
|
|
1 |
import gradio as gr
|
2 |
import torch
|
3 |
import json
|
|
|
4 |
from torch import nn
|
5 |
import requests
|
6 |
from pathlib import Path
|
7 |
+
import logging
|
8 |
+
|
9 |
+
# Set up logging
|
10 |
+
logging.basicConfig(level=logging.INFO)
|
11 |
+
logger = logging.getLogger(__name__)
|
12 |
|
13 |
class TextGenerator(nn.Module):
|
14 |
+
def __init__(self, vocab_size, embedding_dim=256, hidden_dim=512):
|
15 |
super().__init__()
|
16 |
self.embedding = nn.Embedding(vocab_size, embedding_dim)
|
17 |
self.lstm = nn.LSTM(embedding_dim, hidden_dim, batch_first=True)
|
|
|
23 |
return self.fc(lstm_out)
|
24 |
|
25 |
def download_file(url, local_path):
|
26 |
+
try:
|
27 |
+
response = requests.get(url)
|
28 |
+
response.raise_for_status() # Raise an exception for bad status codes
|
29 |
Path(local_path).parent.mkdir(parents=True, exist_ok=True)
|
30 |
with open(local_path, 'wb') as f:
|
31 |
f.write(response.content)
|
32 |
+
logger.info(f"Successfully downloaded {url} to {local_path}")
|
33 |
+
except Exception as e:
|
34 |
+
logger.error(f"Error downloading {url}: {str(e)}")
|
35 |
+
raise
|
36 |
|
37 |
def load_model_and_tokenizers():
|
38 |
# Create a local directory for downloaded files
|
39 |
cache_dir = Path("model_cache")
|
40 |
cache_dir.mkdir(exist_ok=True)
|
41 |
|
42 |
+
# Default configuration values
|
43 |
+
default_config = {
|
44 |
+
'vocab_size': 10000, # Default vocabulary size
|
45 |
+
'embedding_dim': 256, # Default embedding dimension
|
46 |
+
'hidden_dim': 512 # Default hidden dimension
|
47 |
+
}
|
48 |
+
|
49 |
# URLs for the files
|
50 |
base_url = "https://huggingface.co/Omarrran/temp_data/raw/main"
|
51 |
files = {
|
|
|
59 |
for filename, url in files.items():
|
60 |
local_path = cache_dir / filename
|
61 |
if not local_path.exists():
|
62 |
+
logger.info(f"Downloading {filename}...")
|
63 |
download_file(url, local_path)
|
64 |
|
65 |
+
try:
|
66 |
+
# Load configuration
|
67 |
+
with open(cache_dir / "model_config.json", "r") as f:
|
68 |
+
config = json.load(f)
|
69 |
+
# Merge with default config
|
70 |
+
for key in default_config:
|
71 |
+
if key not in config:
|
72 |
+
logger.warning(f"Configuration parameter '{key}' not found, using default value: {default_config[key]}")
|
73 |
+
config[key] = default_config[key]
|
74 |
+
except Exception as e:
|
75 |
+
logger.warning(f"Error loading config file: {str(e)}. Using default configuration.")
|
76 |
+
config = default_config
|
77 |
|
78 |
+
try:
|
79 |
+
# Load tokenizers
|
80 |
+
with open(cache_dir / "word_to_int.json", "r") as f:
|
81 |
+
word_to_int = json.load(f)
|
82 |
+
with open(cache_dir / "int_to_word.json", "r") as f:
|
83 |
+
int_to_word = json.load(f)
|
84 |
+
|
85 |
+
# Update vocab size based on actual vocabulary
|
86 |
+
config['vocab_size'] = len(word_to_int)
|
87 |
+
|
88 |
+
except Exception as e:
|
89 |
+
logger.error(f"Error loading tokenizer files: {str(e)}")
|
90 |
+
raise
|
|
|
|
|
|
|
91 |
|
92 |
+
try:
|
93 |
+
# Initialize model
|
94 |
+
model = TextGenerator(
|
95 |
+
vocab_size=config['vocab_size'],
|
96 |
+
embedding_dim=config['embedding_dim'],
|
97 |
+
hidden_dim=config['hidden_dim']
|
98 |
+
)
|
99 |
+
|
100 |
+
# Load model weights
|
101 |
+
model.load_state_dict(torch.load(cache_dir / "model.pt", map_location=torch.device('cpu')))
|
102 |
+
model.eval()
|
103 |
+
|
104 |
+
return model, word_to_int, int_to_word
|
105 |
+
|
106 |
+
except Exception as e:
|
107 |
+
logger.error(f"Error loading model: {str(e)}")
|
108 |
+
raise
|
109 |
|
110 |
def generate_text(prompt, max_length=100):
|
111 |
+
try:
|
112 |
+
# Load model and tokenizers
|
113 |
+
model, word_to_int, int_to_word = load_model_and_tokenizers()
|
114 |
+
|
115 |
+
# Tokenize input prompt
|
116 |
+
input_ids = [word_to_int.get(word, word_to_int.get('<UNK>', 0)) for word in prompt.split()]
|
117 |
+
input_tensor = torch.tensor([input_ids])
|
118 |
+
|
119 |
+
# Generate text
|
120 |
+
generated_ids = input_ids.copy()
|
121 |
+
|
122 |
+
with torch.no_grad():
|
123 |
+
for _ in range(max_length):
|
124 |
+
current_input = torch.tensor([generated_ids[-50:]]) # Use last 50 tokens as context
|
125 |
+
outputs = model(current_input)
|
126 |
+
next_token_id = outputs[0, -1, :].argmax().item()
|
127 |
+
generated_ids.append(next_token_id)
|
128 |
+
|
129 |
+
if next_token_id == word_to_int.get('<EOS>', 0):
|
130 |
+
break
|
131 |
+
|
132 |
+
# Convert ids back to text
|
133 |
+
generated_text = ' '.join([int_to_word.get(str(idx), '<UNK>') for idx in generated_ids])
|
134 |
+
return generated_text
|
135 |
+
|
136 |
+
except Exception as e:
|
137 |
+
logger.error(f"Error in text generation: {str(e)}")
|
138 |
+
return f"Error generating text: {str(e)}"
|
139 |
|
140 |
# Create Gradio interface
|
141 |
iface = gr.Interface(
|