Spaces:
Running
Running
File size: 9,055 Bytes
c2ea5ed 7bc750c c2ea5ed 7bc750c |
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 |
"""
Unified Knowledge Extraction Method (1-Task Approach)
Copied from core/agent_monitoring_unified.py and adapted for evaluation framework.
Uses the unified 1-task CrewAI approach with a single agent that performs all
knowledge extraction tasks in one step.
"""
# Import the LiteLLM fix FIRST, before any other imports that might use LiteLLM
import os
import sys
# Add the parent directory to the path to ensure imports work correctly
sys.path.append(os.path.dirname(os.path.dirname(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))))
import json
import logging
import time
from datetime import datetime
from typing import Any, Dict
from crewai import Agent, Crew, Process, Task
from evaluation.knowledge_extraction.baselines.base_method import BaseKnowledgeExtractionMethod
from evaluation.knowledge_extraction.utils.models import KnowledgeGraph
# Import shared prompt templates
from evaluation.knowledge_extraction.utils.prompts import (
ENTITY_EXTRACTION_INSTRUCTION_PROMPT,
ENTITY_EXTRACTION_SYSTEM_PROMPT,
GRAPH_BUILDER_SYSTEM_PROMPT,
RELATION_EXTRACTION_INSTRUCTION_PROMPT,
RELATION_EXTRACTION_SYSTEM_PROMPT,
)
from utils.fix_litellm_stop_param import * # This applies the patches # noqa: F403
# Configure logging
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)
# Set higher log levels for noisy libraries
logging.getLogger("openai").setLevel(logging.WARNING)
logging.getLogger("httpx").setLevel(logging.WARNING)
logging.getLogger("litellm").setLevel(logging.WARNING)
logging.getLogger("chromadb").setLevel(logging.WARNING)
# Set default verbosity level
verbose_level = 0
# Set environment variables
os.environ["OPENAI_MODEL_NAME"] = "gpt-5-mini"
class UnifiedKnowledgeExtractionMethod(BaseKnowledgeExtractionMethod):
"""Unified 1-task knowledge extraction method using CrewAI."""
def __init__(self, **kwargs):
super().__init__("unified_method", **kwargs)
self._setup_agent_and_task()
def _setup_agent_and_task(self):
"""Set up the CrewAI agent and task."""
# Create unified agent
self.unified_knowledge_graph_agent = Agent(
role="Unified Knowledge Graph Analyst",
goal="Create comprehensive knowledge graphs from agent system data in a single analysis pass",
backstory=f"""{ENTITY_EXTRACTION_SYSTEM_PROMPT}
{RELATION_EXTRACTION_SYSTEM_PROMPT}
{GRAPH_BUILDER_SYSTEM_PROMPT}.""",
verbose=bool(verbose_level),
llm=os.environ["OPENAI_MODEL_NAME"]
)
# Create unified task
self.unified_knowledge_graph_task = Task(
description=f"""
Extract entities:
{ENTITY_EXTRACTION_INSTRUCTION_PROMPT}
Also extract relationships:
{RELATION_EXTRACTION_INSTRUCTION_PROMPT}
Finally, build the knowledge graph:
""",
agent=self.unified_knowledge_graph_agent,
expected_output="A complete knowledge graph with entities, relations, and metadata",
output_pydantic=KnowledgeGraph,
)
# Create crew
self.unified_agent_monitoring_crew = Crew(
agents=[self.unified_knowledge_graph_agent],
tasks=[self.unified_knowledge_graph_task],
verbose=bool(verbose_level),
memory=False,
planning=False,
process=Process.sequential,
)
def process_text(self, text: str) -> Dict[str, Any]:
"""
Process input text using the unified 1-task CrewAI approach.
Args:
text: Input text to process
Returns:
Dictionary with kg_data, metadata, success, and optional error
"""
start_time = time.time()
try:
logger.info(f"process_text called with text length: {len(text)}")
logger.info(f"text first 200 chars: {repr(text[:200])}")
logger.info("Starting crew execution with input_data...")
# Run the crew with proper input mechanism
result = self.unified_agent_monitoring_crew.kickoff(inputs={"input_data": text})
logger.info(f"Crew execution completed, result type: {type(result)}")
processing_time = time.time() - start_time
# Extract the knowledge graph from the result
if hasattr(result, 'pydantic') and result.pydantic:
kg_data = result.pydantic.dict()
elif hasattr(result, 'raw'):
# Try to parse as JSON
try:
kg_data = json.loads(result.raw)
except: # noqa: E722
kg_data = {"entities": [], "relations": [], "error": "Failed to parse result"}
else:
kg_data = {"entities": [], "relations": [], "error": "Unknown result format"}
# Validate kg_data structure
if not isinstance(kg_data, dict):
raise ValueError("kg_data is not a dict after parsing")
if not ("entities" in kg_data and "relations" in kg_data):
raise ValueError("kg_data missing 'entities' or 'relations'")
# Add metadata
if "metadata" not in kg_data:
kg_data["metadata"] = {}
kg_data["metadata"]["processing_info"] = {
"method": "unified_single_task",
"processing_time_seconds": processing_time,
"processed_at": datetime.now().isoformat(),
"agent_count": 1,
"task_count": 1,
"api_calls": 1
}
# Calculate statistics
entity_count = len(kg_data.get("entities", []))
relation_count = len(kg_data.get("relations", []))
return {
"success": True,
"kg_data": kg_data,
"metadata": {
"approach": "unified_1_task",
"tasks_executed": 1,
"agents_used": 1,
"method": self.method_name,
"processing_time_seconds": processing_time,
"entity_count": entity_count,
"relation_count": relation_count,
"entities_per_second": entity_count / processing_time if processing_time > 0 else 0,
"relations_per_second": relation_count / processing_time if processing_time > 0 else 0,
"api_calls": 1
}
}
except Exception as e:
processing_time = time.time() - start_time
logger.error(f"Error in unified knowledge extraction method: {e}")
logger.error(f"Error type: {type(e).__name__}")
import traceback
logger.error(f"Traceback: {traceback.format_exc()}")
return {
"success": False,
"error": str(e),
"kg_data": {"entities": [], "relations": []},
"metadata": {
"approach": "unified_1_task",
"tasks_executed": 0,
"agents_used": 0,
"method": self.method_name,
"processing_time_seconds": processing_time,
"api_calls": 1
}
}
def extract_knowledge_graph(self, trace_data: str) -> Dict[str, Any]:
"""
Extract knowledge graph from trace data.
Args:
trace_data: Agent trace data as JSON string
Returns:
Dictionary with entities and relations
"""
try:
# Debug logging
logger.info(f"extract_knowledge_graph called with trace_data type: {type(trace_data)}")
if isinstance(trace_data, str):
logger.info(f"trace_data length: {len(trace_data)}")
logger.info(f"trace_data first 200 chars: {repr(trace_data[:200])}")
# Pass the JSON string directly to process_text without re-encoding
result = self.process_text(trace_data)
# Return just the knowledge graph data
if result.get("success", False):
return result.get("kg_data", {"entities": [], "relations": []})
else:
# Return empty knowledge graph on failure
return {"entities": [], "relations": []}
except Exception as e:
logger.error(f"Error in extract_knowledge_graph: {e}")
logger.error(f"trace_data type: {type(trace_data)}")
if isinstance(trace_data, str):
logger.error(f"trace_data content (first 200 chars): {repr(trace_data[:200])}")
return {"entities": [], "relations": []}
|