File size: 8,145 Bytes
dcc24f8 |
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 |
"""
Inference/Prediction Module
Load fine-tuned model and extract entities from emails.
"""
import json
import re
from pathlib import Path
from typing import Dict, Optional, Union
from dataclasses import dataclass
@dataclass
class PredictionResult:
"""Result of model prediction."""
entities: Dict
raw_response: str
success: bool
error: Optional[str] = None
def to_dict(self) -> Dict:
return {
"entities": self.entities,
"raw_response": self.raw_response,
"success": self.success,
"error": self.error
}
def to_json(self, indent: int = 2) -> str:
return json.dumps(self.entities, indent=indent)
class Predictor:
"""
Load and run inference with fine-tuned model.
Supports both:
- Base model + LoRA adapter
- Merged model
"""
EXTRACTION_PROMPT = """Extract financial entities from this email:
Subject: {subject}
Body: {body}"""
def __init__(
self,
model_path: Union[str, Path],
adapter_path: Optional[Union[str, Path]] = None,
max_tokens: int = 200
):
"""
Initialize predictor.
Args:
model_path: Path to model (base or merged)
adapter_path: Optional path to LoRA adapter
max_tokens: Maximum tokens to generate
"""
self.model_path = Path(model_path)
self.adapter_path = Path(adapter_path) if adapter_path else None
self.max_tokens = max_tokens
self.model = None
self.tokenizer = None
self._loaded = False
def load(self):
"""Load the model and tokenizer."""
if self._loaded:
return
try:
from mlx_lm import load
except ImportError:
raise ImportError(
"mlx_lm is required. Install with: pip install mlx-lm"
)
print(f"🔄 Loading model from {self.model_path}...")
if self.adapter_path:
print(f" With adapter: {self.adapter_path}")
self.model, self.tokenizer = load(
str(self.model_path),
adapter_path=str(self.adapter_path)
)
else:
self.model, self.tokenizer = load(str(self.model_path))
self._loaded = True
print("✅ Model loaded successfully!")
def predict(
self,
subject: str = "",
body: str = "",
email_text: Optional[str] = None
) -> PredictionResult:
"""
Extract entities from an email.
Args:
subject: Email subject
body: Email body
email_text: Full email text (alternative to subject+body)
Returns:
PredictionResult with extracted entities
"""
if not self._loaded:
self.load()
try:
from mlx_lm import generate
except ImportError:
raise ImportError("mlx_lm is required")
# Build prompt
if email_text:
prompt = f"Extract financial entities from this email:\n\n{email_text}"
else:
prompt = self.EXTRACTION_PROMPT.format(
subject=subject[:200],
body=body[:1500]
)
# Generate response
try:
response = generate(
self.model,
self.tokenizer,
prompt=prompt,
max_tokens=self.max_tokens,
verbose=False
)
except Exception as e:
return PredictionResult(
entities={},
raw_response="",
success=False,
error=f"Generation failed: {str(e)}"
)
# Parse JSON from response
entities = self._extract_json(response)
return PredictionResult(
entities=entities if entities else {},
raw_response=response,
success=entities is not None
)
def predict_batch(
self,
emails: list
) -> list:
"""
Extract entities from multiple emails.
Args:
emails: List of dicts with 'subject' and 'body' keys
Returns:
List of PredictionResults
"""
results = []
for email in emails:
result = self.predict(
subject=email.get('subject', ''),
body=email.get('body', '')
)
results.append(result)
return results
def _extract_json(self, response: str) -> Optional[Dict]:
"""Extract JSON object from model response."""
# Find JSON pattern
match = re.search(r'\{[^{}]*\}', response)
if match:
try:
return json.loads(match.group())
except json.JSONDecodeError:
pass
return None
def main():
"""CLI for running predictions."""
import argparse
parser = argparse.ArgumentParser(
description="Extract financial entities from emails using fine-tuned LLM"
)
parser.add_argument(
"--model",
type=str,
default=str(Path.home() / "llm-mail-trainer/models/base/phi3-mini"),
help="Path to model"
)
parser.add_argument(
"--adapter",
type=str,
default=None,
help="Path to LoRA adapter (optional)"
)
parser.add_argument(
"--subject",
type=str,
default="",
help="Email subject"
)
parser.add_argument(
"--body",
type=str,
default=None,
help="Email body text"
)
parser.add_argument(
"--file",
type=str,
default=None,
help="Path to file containing email text"
)
parser.add_argument(
"--interactive",
action="store_true",
help="Run in interactive mode"
)
args = parser.parse_args()
# Initialize predictor
predictor = Predictor(
model_path=args.model,
adapter_path=args.adapter
)
if args.interactive:
run_interactive(predictor)
elif args.file:
with open(args.file, 'r') as f:
text = f.read()
predictor.load()
result = predictor.predict(email_text=text)
print(result.to_json())
elif args.body:
predictor.load()
result = predictor.predict(subject=args.subject, body=args.body)
print(result.to_json())
else:
parser.print_help()
def run_interactive(predictor: Predictor):
"""Interactive mode for testing."""
predictor.load()
print("\n" + "=" * 60)
print("🧠 LLM Mail Trainer - Interactive Mode")
print("=" * 60)
print("Enter email text to extract entities.")
print("Type 'quit' or 'exit' to stop.")
print("=" * 60 + "\n")
while True:
print("\n📧 Enter email text (multi-line, end with empty line):")
lines = []
while True:
try:
line = input()
if line.lower() in ['quit', 'exit']:
print("\n👋 Goodbye!")
return
if line == "" and lines:
break
lines.append(line)
except EOFError:
print("\n👋 Goodbye!")
return
email_text = "\n".join(lines)
if email_text.strip():
print("\n🔄 Extracting entities...")
result = predictor.predict(email_text=email_text)
print("\n📋 Extracted Entities:")
print("-" * 40)
print(result.to_json())
if not result.success:
print(f"\n⚠️ Warning: {result.error or 'Could not parse JSON from response'}")
print(f"Raw response: {result.raw_response[:200]}...")
if __name__ == "__main__":
main()
|