Spaces:
Sleeping
Sleeping
File size: 10,445 Bytes
2cb327c | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 | """
Summarization Model Configuration
===================================
Manages AI model selection and inference for article summarization.
English Pipeline:
Uses configurable models (default: t5-small) via MODEL_TYPE env var.
Options: t5-small, distilbart, bart, t5, pegasus, led
Hindi Pipeline (via english_summary.py fallback only):
Uses L3Cube-Pune/Hindi-BART-Summary.
NOTE: The recommended Hindi path is hindi_summary.py (mT5 ONNX + Groq).
This model is only used if summarize is called directly on Hindi articles.
Architecture:
SummarizationModel is a thread-safe singleton. Each language's model is
loaded lazily on first use and cached in memory for the batch run.
Usage:
from backend.summarization.model import get_summarizer
summarizer = get_summarizer()
summary = summarizer.summarize(article_text, max_words=150, language="english")
"""
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
import torch
import os
import threading
from dotenv import load_dotenv
load_dotenv()
# βββββββββββββββββββββββββββββββββββββββββββββ
# Device Configuration
# βββββββββββββββββββββββββββββββββββββββββββββ
ENV_DEVICE = os.getenv('DEVICE', 'cpu').lower()
if ENV_DEVICE == 'gpu':
USE_GPU = True
DEVICE = 0 if torch.cuda.is_available() else -1
if DEVICE == -1:
print("Warning: GPU requested but CUDA not available, falling back to CPU")
elif ENV_DEVICE == 'cpu':
USE_GPU = False
DEVICE = -1
else:
USE_GPU = True
DEVICE = 0 if torch.cuda.is_available() else -1
# CPU Optimization
if DEVICE == -1:
CPU_THREADS = int(os.getenv('MAX_WORKERS', '4'))
torch.set_num_threads(CPU_THREADS)
torch.set_num_interop_threads(CPU_THREADS)
# βββββββββββββββββββββββββββββββββββββββββββββ
# Model Selection
# βββββββββββββββββββββββββββββββββββββββββββββ
MODEL_TYPE = os.getenv('MODEL_TYPE', 't5-small').lower()
HINDI_MODEL_NAME = os.getenv('HINDI_MODEL_NAME', 'L3Cube-Pune/Hindi-BART-Summary')
MODELS = {
"t5-small": {
"name": "t5-small",
"max_length": 300,
"min_length": 80,
"max_input_length": 1024,
"description": "T5 Small - Fast CPU inference, ~240MB (BEST FOR GITHUB ACTIONS)"
},
"distilbart": {
"name": "sshleifer/distilbart-cnn-12-6",
"max_length": 130,
"min_length": 30,
"max_input_length": 1024,
"description": "DistilBART - Faster than BART, ~600MB (GOOD FOR GITHUB ACTIONS)"
},
"bart": {
"name": "facebook/bart-large-cnn",
"max_length": 130,
"min_length": 30,
"max_input_length": 1024,
"description": "BART - Good balance of speed and quality, ~1.6GB"
},
"t5": {
"name": "t5-base",
"max_length": 150,
"min_length": 30,
"max_input_length": 512,
"description": "T5 Base - Versatile text-to-text model, ~850MB"
},
"pegasus": {
"name": "google/pegasus-xsum",
"max_length": 128,
"min_length": 32,
"max_input_length": 512,
"description": "Pegasus - Optimized for news summarization, ~2.2GB"
},
"led": {
"name": "allenai/led-base-16384",
"max_length": 150,
"min_length": 30,
"max_input_length": 4096,
"description": "LED - Best for long documents, ~500MB"
}
}
if MODEL_TYPE not in MODELS:
valid_models = ", ".join(MODELS.keys())
print(f"Warning: Invalid MODEL_TYPE '{MODEL_TYPE}' in .env")
print(f"Valid options: {valid_models}")
print(f"Falling back to default: t5-small")
MODEL_TYPE = "t5-small"
LANGUAGE_MODEL_OVERRIDES = {
"hindi": {
"name": HINDI_MODEL_NAME,
"max_length": 220,
"min_length": 70,
"max_input_length": 1024,
"description": "Hindi-BART-Summary by L3Cube Pune",
"is_t5": False
}
}
def _fallback_summary(words, max_words: int) -> str:
return " ".join(words[:max_words]).strip()
def _normalize_summary_length(summary: str, original_words, max_words: int) -> str:
if not summary:
return _fallback_summary(original_words, max_words)
summary_words = summary.split()
if len(summary_words) > max_words:
summary = " ".join(summary_words[:max_words]).strip()
summary_words = summary.split()
min_words = max(35, int(max_words * 0.55))
if len(summary_words) < min_words:
return _fallback_summary(original_words, max_words)
return summary
def _language_model_config(language: str):
lang = (language or "english").strip().lower()
if lang in LANGUAGE_MODEL_OVERRIDES:
return LANGUAGE_MODEL_OVERRIDES[lang], lang
return MODELS[MODEL_TYPE], "english"
def _is_t5_model(language: str) -> bool:
lang = (language or "english").strip().lower()
if lang in LANGUAGE_MODEL_OVERRIDES:
return LANGUAGE_MODEL_OVERRIDES[lang].get("is_t5", False)
return MODEL_TYPE.startswith("t5")
class SummarizationModel:
"""Thread-safe singleton for loading and running HuggingFace seq2seq models.
Lazy-loads the model on first summarize() call for each language.
Protects tokenization/generation with a threading lock.
"""
_instance = None
def __new__(cls):
if cls._instance is None:
cls._instance = super().__new__(cls)
cls._instance._lock = threading.Lock()
cls._instance._models = {}
return cls._instance
def __init__(self):
if not hasattr(self, "_models"):
self._models = {}
def _load_model(self, language: str):
model_config, model_key = _language_model_config(language)
device_name = "GPU (CUDA)" if DEVICE == 0 else "CPU"
if DEVICE == 0 and torch.cuda.is_available():
gpu_name = torch.cuda.get_device_name(0)
print(f"Using device: {device_name} ({gpu_name})")
else:
print(f"Using device: {device_name}")
if USE_GPU and not torch.cuda.is_available():
print("Warning: GPU requested but CUDA not available, falling back to CPU")
print(f"Loading model: {model_config['name']}")
print(f"Description: {model_config['description']}")
try:
tokenizer = AutoTokenizer.from_pretrained(model_config["name"])
model = AutoModelForSeq2SeqLM.from_pretrained(model_config["name"])
if DEVICE == 0:
model = model.to("cuda")
self._models[model_key] = {
"tokenizer": tokenizer,
"model": model,
"config": model_config,
"device": "cuda" if DEVICE == 0 else "cpu"
}
print("Model loaded successfully!\n")
except Exception as e:
print(f"Error loading model: {e}")
raise
def summarize(self, text: str, max_words: int = 80, language: str = "english") -> str:
"""Generate a summary of the input text.
Args:
text: The article body text to summarize.
max_words: Maximum number of words in the output summary.
language: "english" or "hindi" β determines which model to use.
Returns:
Summary string. Falls back to truncated original if model fails.
"""
if not text or not text.strip():
return text
words = text.split()
max_input_words = 600
if len(words) > max_input_words:
text = " ".join(words[:max_input_words])
if len(words) < 40:
return text
model_config, model_key = _language_model_config(language)
if model_key not in self._models:
with self._lock:
if model_key not in self._models:
self._load_model(model_key)
model_bundle = self._models[model_key]
tokenizer = model_bundle["tokenizer"]
model = model_bundle["model"]
device = model_bundle["device"]
if _is_t5_model(model_key):
text = "summarize: " + text
max_length = min(int(max_words * 2.0), model_config["max_length"])
min_length = min(max(model_config["min_length"], int(max_words * 0.5)), max_length - 20)
min_length = max(20, min_length)
try:
with self._lock:
inputs = tokenizer(
text,
max_length=model_config["max_input_length"],
truncation=True,
return_tensors="pt"
)
if device == "cuda":
inputs = {k: v.to("cuda") for k, v in inputs.items()}
if _is_t5_model(model_key):
summary_ids = model.generate(
inputs["input_ids"],
max_length=max_length,
min_length=min_length,
num_beams=4,
length_penalty=2.5,
early_stopping=True,
no_repeat_ngram_size=3
)
else:
summary_ids = model.generate(
inputs["input_ids"],
max_length=max_length,
min_length=min_length,
num_beams=4,
length_penalty=2.0,
early_stopping=True
)
summary = tokenizer.decode(
summary_ids[0],
skip_special_tokens=True,
clean_up_tokenization_spaces=True
)
if not summary or len(summary.strip()) < 20:
return _fallback_summary(words, max_words)
return _normalize_summary_length(summary.strip(), words, max_words)
except Exception as e:
print(f"Summarization error: {e}")
return _fallback_summary(words, max_words)
def get_summarizer():
"""Returns the singleton SummarizationModel instance."""
return SummarizationModel()
|