Spaces:
Sleeping
Sleeping
File size: 9,118 Bytes
5dbca28 | 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 | """
Pipeline Orchestrator
=====================
Ties all stages together into a single callable pipeline:
ingest → clean → segment → qa_detect → filter → extract → normalize → validate → JSON
Accepts single contract or batch processing.
Logs intermediate outputs at each stage for debugging.
"""
import json
import logging
import os
from typing import Any, Dict, List, Optional, Union
from all_model_code.model_1_code.stage1_ingestion import ingest
from all_model_code.model_1_code.stage2_cleaning import clean_text
from all_model_code.model_1_code.stage3_segmentation import Chunk, segment_text
from all_model_code.model_1_code.stage4_qa_detection import QAClauseDetector, OBLIGATION_QUESTIONS
from all_model_code.model_1_code.stage5_span_filter import filter_spans
from all_model_code.model_1_code.stage6_extraction import extract_obligations
from all_model_code.model_1_code.stage7_normalize import normalize_and_validate
from all_model_code.model_1_code.utils import get_device
logger = logging.getLogger(__name__)
# Default configuration
DEFAULT_CONFIG = {
# QA Model
"model_name": "deepset/roberta-base-squad2",
"device": "auto", # auto-detect GPU (RTX 4050 6GB VRAM)
# Segmentation
"max_chunk_chars": 1500,
"min_chunk_chars": 50,
"overlap_chars": 200,
# QA Detection
"qa_confidence_threshold": 0.01,
"max_answer_length": 200,
# Span Filtering
"filter_min_confidence": 0.05,
"filter_min_length": 10,
"filter_max_length": 2000,
"filter_deduplicate": True,
# Normalization & Validation
"require_metric": False,
"require_value": False,
"min_fields": 2,
}
class ObligationPipeline:
"""End-to-end contract obligation extraction pipeline.
Usage:
pipeline = ObligationPipeline()
results = pipeline.process("path/to/contract.pdf")
# or
results = pipeline.process(raw_text, source_type="text")
"""
def __init__(self, config: Optional[Dict[str, Any]] = None):
"""Initialize pipeline with configuration.
Args:
config: Optional config dict. Missing keys use DEFAULT_CONFIG.
"""
self.config = {**DEFAULT_CONFIG, **(config or {})}
self.detector = None # Lazy-loaded
# Stage-level intermediate results (for debugging)
self._debug = {}
def _get_detector(self) -> QAClauseDetector:
"""Get or create the QA detector (lazy loading)."""
if self.detector is None:
self.detector = QAClauseDetector(
model_name=self.config["model_name"],
device=self.config["device"],
confidence_threshold=self.config["qa_confidence_threshold"],
max_answer_length=self.config["max_answer_length"],
)
return self.detector
def process(
self,
source: str,
source_type: Optional[str] = None,
contract_id: Optional[str] = None,
questions: Optional[dict] = None,
debug: bool = False,
) -> List[Dict[str, Any]]:
"""Run the full pipeline on a single contract.
Args:
source: PDF path or raw text.
source_type: 'pdf' or 'text' (auto-detected if None).
contract_id: Optional identifier for the contract.
questions: Optional custom questions dict.
debug: If True, store intermediate results in self._debug.
Returns:
List of structured obligation dicts ready for Model 2.
"""
self._debug = {}
# ─── Stage 1: Ingestion ──────────────────────────────────────────
logger.info("Stage 1: Ingestion")
raw_text = ingest(source, source_type)
if debug:
self._debug["raw_text_length"] = len(raw_text)
# ─── Stage 2: Text Cleaning ──────────────────────────────────────
logger.info("Stage 2: Text Cleaning")
cleaned = clean_text(raw_text)
if debug:
self._debug["cleaned_text_length"] = len(cleaned)
# ─── Stage 3: Paragraph Segmentation ─────────────────────────────
logger.info("Stage 3: Paragraph Segmentation")
chunks = segment_text(
cleaned,
max_chunk_chars=self.config["max_chunk_chars"],
min_chunk_chars=self.config["min_chunk_chars"],
overlap_chars=self.config["overlap_chars"],
)
logger.info(f" → {len(chunks)} chunks")
if debug:
self._debug["num_chunks"] = len(chunks)
self._debug["chunk_lengths"] = [len(c) for c in chunks]
# ─── Stage 4: QA-Based Clause Detection ──────────────────────────
logger.info("Stage 4: QA-Based Clause Detection")
detector = self._get_detector()
detections = detector.detect_in_chunks(
chunks,
questions=questions or OBLIGATION_QUESTIONS,
show_progress=True,
)
logger.info(f" → {len(detections)} raw detections")
if debug:
self._debug["num_raw_detections"] = len(detections)
# ─── Stage 5: Span Filtering ─────────────────────────────────────
logger.info("Stage 5: Span Filtering")
filtered = filter_spans(
detections,
min_confidence=self.config["filter_min_confidence"],
min_length=self.config["filter_min_length"],
max_length=self.config["filter_max_length"],
deduplicate=self.config["filter_deduplicate"],
)
logger.info(f" → {len(filtered)} filtered spans")
if debug:
self._debug["num_filtered"] = len(filtered)
# ─── Stage 6: Information Extraction ─────────────────────────────
logger.info("Stage 6: Information Extraction")
obligations = extract_obligations(filtered)
if debug:
self._debug["num_extracted"] = len(obligations)
# ─── Stage 7: Normalization & Validation ─────────────────────────
logger.info("Stage 7: Normalization & Validation")
results = normalize_and_validate(
obligations,
require_metric=self.config["require_metric"],
require_value=self.config["require_value"],
min_fields=self.config["min_fields"],
)
logger.info(f" → {len(results)} validated obligations")
# Add contract_id to all results
if contract_id:
for r in results:
r["contract_id"] = contract_id
if debug:
self._debug["num_final"] = len(results)
return results
def process_batch(
self,
sources: List[Dict[str, str]],
debug: bool = False,
) -> Dict[str, List[Dict[str, Any]]]:
"""Process multiple contracts.
Args:
sources: List of dicts with keys 'source', 'source_type', 'contract_id'.
debug: If True, store per-contract debug info.
Returns:
Dict mapping contract_id → list of obligation dicts.
"""
all_results = {}
for i, src_info in enumerate(sources):
cid = src_info.get("contract_id", f"contract_{i}")
logger.info(f"\n{'='*60}")
logger.info(f"Processing contract {i+1}/{len(sources)}: {cid}")
logger.info(f"{'='*60}")
results = self.process(
source=src_info["source"],
source_type=src_info.get("source_type"),
contract_id=cid,
debug=debug,
)
all_results[cid] = results
# Summary
total = sum(len(v) for v in all_results.values())
logger.info(f"\nBatch complete: {total} obligations from {len(sources)} contracts")
return all_results
def get_debug_info(self) -> Dict[str, Any]:
"""Get intermediate results from the last process() call."""
return self._debug
def save_results(results: Union[List, Dict], output_path: str):
"""Save pipeline results to a JSON file.
Args:
results: Obligation dicts from process() or process_batch().
output_path: Path to output JSON file.
"""
os.makedirs(os.path.dirname(output_path) or ".", exist_ok=True)
with open(output_path, "w", encoding="utf-8") as f:
json.dump(results, f, indent=2, ensure_ascii=False)
logger.info(f"Results saved to {output_path}")
|