File size: 20,251 Bytes
5e4510c |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 |
"""
Evolution trace logging for OpenEvolve
This module provides functionality to log detailed traces of program evolution,
capturing state-action-reward transitions for RL training and analysis.
"""
import json
import logging
import time
from dataclasses import asdict, dataclass
from pathlib import Path
from typing import Any, Dict, List, Optional, Union
from openevolve.utils.trace_export_utils import (
append_trace_jsonl,
export_traces,
export_traces_json,
)
logger = logging.getLogger(__name__)
@dataclass
class EvolutionTrace:
"""Represents a single evolution trace entry"""
iteration: int
timestamp: float
parent_id: str
child_id: str
parent_metrics: Dict[str, Any]
child_metrics: Dict[str, Any]
parent_code: Optional[str] = None
child_code: Optional[str] = None
code_diff: Optional[str] = None
prompt: Optional[Dict[str, str]] = None
llm_response: Optional[str] = None
improvement_delta: Optional[Dict[str, float]] = None
island_id: Optional[int] = None
generation: Optional[int] = None
artifacts: Optional[Dict[str, Any]] = None
metadata: Optional[Dict[str, Any]] = None
def to_dict(self) -> Dict[str, Any]:
"""Convert trace to dictionary format"""
return {k: v for k, v in asdict(self).items() if v is not None}
def calculate_improvement(self) -> Dict[str, float]:
"""Calculate improvement deltas between parent and child metrics"""
improvement = {}
for key in self.child_metrics:
if key in self.parent_metrics:
parent_val = self.parent_metrics[key]
child_val = self.child_metrics[key]
if isinstance(parent_val, (int, float)) and isinstance(child_val, (int, float)):
improvement[key] = child_val - parent_val
return improvement
class EvolutionTracer:
"""Manages evolution trace logging with support for multiple formats"""
def __init__(
self,
output_path: Optional[str] = None,
format: str = "jsonl",
include_code: bool = False,
include_prompts: bool = True,
enabled: bool = True,
buffer_size: int = 10,
compress: bool = False,
):
"""
Initialize the evolution tracer
Args:
output_path: Path to save trace data
format: Output format ('jsonl', 'json', 'hdf5')
include_code: Whether to include full code in traces
include_prompts: Whether to include prompts and LLM responses
enabled: Whether tracing is enabled
buffer_size: Number of traces to buffer before writing
compress: Whether to compress output files
"""
self.enabled = enabled
self.format = format
self.include_code = include_code
self.include_prompts = include_prompts
self.compress = compress
self.buffer_size = buffer_size
# Track statistics
self.stats = {
"total_traces": 0,
"improvement_count": 0,
"total_improvement": {},
"best_improvement": {},
"worst_decline": {},
}
if not self.enabled:
logger.info("Evolution tracing is disabled")
return
# Set up output path
if output_path:
self.output_path = Path(output_path)
else:
self.output_path = Path(f"evolution_trace.{format}")
# Add compression extension if needed
if self.compress and format == "jsonl":
self.output_path = self.output_path.with_suffix(".jsonl.gz")
# Create parent directory if needed
self.output_path.parent.mkdir(parents=True, exist_ok=True)
# Initialize buffer for batched writing
self.buffer: List[EvolutionTrace] = []
# For JSON format, keep all traces in memory
if format == "json":
self.json_traces = []
logger.info(f"Evolution tracer initialized: {self.output_path}")
def log_trace(
self,
iteration: int,
parent_program: Any,
child_program: Any,
prompt: Optional[Dict[str, str]] = None,
llm_response: Optional[str] = None,
artifacts: Optional[Dict[str, Any]] = None,
island_id: Optional[int] = None,
metadata: Optional[Dict[str, Any]] = None,
) -> None:
"""
Log an evolution trace entry
Args:
iteration: Current iteration number
parent_program: Parent program object
child_program: Child program object
prompt: Prompt used for evolution
llm_response: LLM response
artifacts: Any artifacts from evaluation
island_id: Island ID if using island-based evolution
metadata: Additional metadata
"""
if not self.enabled:
return
try:
# Create trace entry
trace = EvolutionTrace(
iteration=iteration,
timestamp=time.time(),
parent_id=parent_program.id,
child_id=child_program.id,
parent_metrics=parent_program.metrics,
child_metrics=child_program.metrics,
island_id=island_id,
generation=child_program.generation,
artifacts=artifacts,
metadata=metadata,
)
# Optionally include code
if self.include_code:
trace.parent_code = parent_program.code
trace.child_code = child_program.code
# Optionally include prompts
if self.include_prompts:
trace.prompt = prompt
trace.llm_response = llm_response
# Calculate improvement
trace.improvement_delta = trace.calculate_improvement()
# Update statistics
self._update_stats(trace)
# Add to buffer
self.buffer.append(trace)
# For JSON format, also keep in memory
if self.format == "json":
self.json_traces.append(trace)
# Write if buffer is full
if len(self.buffer) >= self.buffer_size:
self.flush()
except Exception as e:
logger.error(f"Error logging evolution trace: {e}")
def _update_stats(self, trace: EvolutionTrace):
"""Update running statistics"""
self.stats["total_traces"] += 1
if trace.improvement_delta:
# Check if there's improvement in combined_score
if "combined_score" in trace.improvement_delta:
delta = trace.improvement_delta["combined_score"]
if delta > 0:
self.stats["improvement_count"] += 1
# Track cumulative improvements
for metric, delta in trace.improvement_delta.items():
if metric not in self.stats["total_improvement"]:
self.stats["total_improvement"][metric] = 0
self.stats["best_improvement"][metric] = delta
self.stats["worst_decline"][metric] = delta
self.stats["total_improvement"][metric] += delta
if delta > self.stats["best_improvement"][metric]:
self.stats["best_improvement"][metric] = delta
if delta < self.stats["worst_decline"][metric]:
self.stats["worst_decline"][metric] = delta
def flush(self):
"""Write buffered traces to file"""
if not self.enabled or not self.buffer:
return
try:
if self.format == "jsonl":
# Append each trace to the JSONL file
for trace in self.buffer:
append_trace_jsonl(trace, self.output_path, compress=self.compress)
elif self.format == "json":
# For JSON, we'll write everything at close
pass # Traces already added to json_traces
elif self.format == "hdf5":
# For HDF5, we need to write everything at once
# So we'll keep accumulating until close
pass
# Clear buffer after writing (except for formats that need full data)
if self.format == "jsonl":
self.buffer.clear()
except Exception as e:
logger.error(f"Error flushing traces to file: {e}")
def get_statistics(self) -> Dict[str, Any]:
"""Get current tracing statistics"""
return {
**self.stats,
"improvement_rate": (
self.stats["improvement_count"] / self.stats["total_traces"]
if self.stats["total_traces"] > 0
else 0
),
}
def close(self):
"""Close the tracer and flush remaining data"""
if not self.enabled:
return
# Flush any remaining traces
self.flush()
# For JSON and HDF5, write everything at close
if self.format == "json" and hasattr(self, "json_traces"):
metadata = {
"created_at": time.time(),
"include_code": self.include_code,
"include_prompts": self.include_prompts,
"total_traces": len(self.json_traces),
}
export_traces_json(self.json_traces, self.output_path, metadata=metadata)
elif self.format == "hdf5":
# Export all buffered traces
all_traces = getattr(self, "json_traces", self.buffer)
if all_traces:
metadata = {
"created_at": time.time(),
"include_code": self.include_code,
"include_prompts": self.include_prompts,
}
export_traces(all_traces, self.output_path, format="hdf5", metadata=metadata)
# Log final statistics
stats = self.get_statistics()
logger.info(f"Evolution tracing complete. Total traces: {stats['total_traces']}")
logger.info(f"Improvement rate: {stats['improvement_rate']:.2%}")
if stats["best_improvement"]:
logger.info(f"Best improvements: {stats['best_improvement']}")
if stats["worst_decline"]:
logger.info(f"Worst declines: {stats['worst_decline']}")
def __enter__(self):
"""Context manager entry"""
return self
def __exit__(self, exc_type, exc_val, exc_tb):
"""Context manager exit"""
self.close()
def extract_evolution_trace_from_checkpoint(
checkpoint_dir: Union[str, Path],
output_path: Optional[str] = None,
format: str = "jsonl",
include_code: bool = True,
include_prompts: bool = True,
) -> List[EvolutionTrace]:
"""
Extract evolution traces from an existing checkpoint directory
Args:
checkpoint_dir: Path to checkpoint directory
output_path: Optional path to save extracted traces
format: Output format ('jsonl', 'json')
include_code: Whether to include code in traces
include_prompts: Whether to include prompts in traces
Returns:
List of EvolutionTrace objects
"""
checkpoint_path = Path(checkpoint_dir)
if not checkpoint_path.exists():
raise FileNotFoundError(f"Checkpoint directory {checkpoint_dir} not found")
programs_dir = checkpoint_path / "programs"
if not programs_dir.exists():
raise FileNotFoundError(f"Programs directory not found in {checkpoint_dir}")
logger.info(f"Extracting evolution traces from {checkpoint_dir}")
# Load all programs
programs = {}
program_files = list(programs_dir.glob("*.json"))
for prog_file in program_files:
try:
with open(prog_file, "r") as f:
prog_data = json.load(f)
programs[prog_data["id"]] = prog_data
except (json.JSONDecodeError, KeyError) as e:
logger.warning(f"Error loading program from {prog_file}: {e}")
continue
logger.info(f"Loaded {len(programs)} programs from checkpoint")
# Build parent-child traces
traces = []
for prog_id, prog in programs.items():
# Skip programs without parents
parent_id = prog.get("parent_id")
if not parent_id or parent_id not in programs:
continue
parent = programs[parent_id]
# Create trace entry
trace = EvolutionTrace(
iteration=prog.get("iteration_found", 0),
timestamp=prog.get("timestamp", 0),
parent_id=parent_id,
child_id=prog_id,
parent_metrics=parent.get("metrics", {}),
child_metrics=prog.get("metrics", {}),
generation=prog.get("generation", 0),
island_id=prog.get("metadata", {}).get("island"),
metadata=prog.get("metadata", {}),
)
# Add code if requested
if include_code:
trace.parent_code = parent.get("code", "")
trace.child_code = prog.get("code", "")
# Add prompts if available and requested
if include_prompts:
# Check for prompt data in the program
if "prompts" in prog:
# Prompts might be stored in the program data
trace.prompt = prog["prompts"].get("prompt")
trace.llm_response = prog["prompts"].get("response")
# Calculate improvement
trace.improvement_delta = trace.calculate_improvement()
traces.append(trace)
# Sort traces by iteration
traces.sort(key=lambda x: (x.iteration, x.timestamp))
logger.info(f"Extracted {len(traces)} evolution traces")
# Save to file if output path provided
if output_path:
metadata = {
"total_traces": len(traces),
"extracted_at": time.time(),
"source": "checkpoint",
}
export_traces(traces, output_path, format=format, metadata=metadata)
logger.info(f"Saved {len(traces)} traces to {output_path}")
return traces
def extract_full_lineage_traces(
checkpoint_dir: Union[str, Path], output_path: Optional[str] = None, format: str = "json"
) -> List[Dict[str, Any]]:
"""
Extract complete evolution traces with full lineage chains and prompts/actions
This function builds the complete evolution history for each program,
tracing back through all ancestors to create full lineage chains.
Args:
checkpoint_dir: Path to checkpoint directory
output_path: Optional path to save extracted traces
format: Output format ('json' or 'jsonl')
Returns:
List of lineage trace dictionaries
"""
checkpoint_path = Path(checkpoint_dir)
if not checkpoint_path.exists():
raise FileNotFoundError(f"Checkpoint directory {checkpoint_dir} not found")
programs_dir = checkpoint_path / "programs"
if not programs_dir.exists():
raise FileNotFoundError(f"Programs directory not found in {checkpoint_dir}")
logger.info(f"Extracting full lineage traces from {checkpoint_dir}")
# Load all programs
programs = {}
program_files = list(programs_dir.glob("*.json"))
for prog_file in program_files:
try:
with open(prog_file, "r") as f:
prog_data = json.load(f)
programs[prog_data["id"]] = prog_data
except (json.JSONDecodeError, KeyError) as e:
logger.warning(f"Error loading program from {prog_file}: {e}")
continue
logger.info(f"Loaded {len(programs)} programs from checkpoint")
# Build lineage traces for each program
traces = []
for program_id, program in programs.items():
# Build lineage chain by tracing back through parents
lineage = []
current = program
while current:
lineage.append(current)
parent_id = current.get("parent_id")
current = programs.get(parent_id) if parent_id else None
# Reverse to get chronological order (oldest to newest)
lineage.reverse()
# Extract improvement steps with actions
improvements = []
for i in range(len(lineage) - 1):
parent = lineage[i]
child = lineage[i + 1]
# Extract the action (prompt and response) from child's data
prompts = child.get("prompts", {})
action = None
# Get the prompt used (could be diff_user, full_rewrite_user, or other templates)
for template_key, prompt_data in prompts.items():
action = {
"template": template_key,
"system_prompt": prompt_data.get("system", ""),
"user_prompt": prompt_data.get("user", ""),
"llm_response": (
prompt_data.get("responses", [""])[0]
if prompt_data.get("responses")
else ""
),
}
break # Take the first prompt found
# Calculate improvements for each metric
improvement_deltas = {}
if child.get("metrics") and parent.get("metrics"):
for metric in child["metrics"].keys():
if metric in parent["metrics"]:
parent_val = parent["metrics"][metric]
child_val = child["metrics"][metric]
if isinstance(parent_val, (int, float)) and isinstance(
child_val, (int, float)
):
improvement_deltas[metric] = child_val - parent_val
improvement = {
"step": i,
"parent_id": parent["id"],
"child_id": child["id"],
"parent_metrics": parent.get("metrics", {}),
"child_metrics": child.get("metrics", {}),
"improvement": improvement_deltas,
"generation": child.get("generation", 0),
"iteration_found": child.get("iteration_found", 0),
"changes_description": child.get("metadata", {}).get("changes", ""),
"island_id": child.get("metadata", {}).get("island"),
"action": action, # The prompt/response that led to this improvement
}
improvements.append(improvement)
# Only add traces that have improvement steps
if improvements:
trace = {
"final_program_id": program_id,
"final_metrics": program.get("metrics", {}),
"generation_depth": len(lineage),
"total_iterations": program.get("iteration_found", 0),
"improvement_steps": improvements,
"metadata": {
"language": program.get("language", ""),
"timestamp": program.get("timestamp", 0),
},
}
traces.append(trace)
# Sort traces by generation depth (most evolved first)
traces.sort(key=lambda x: x["generation_depth"], reverse=True)
logger.info(f"Extracted {len(traces)} lineage traces")
# Save to file if output path provided
if output_path:
if format == "json":
metadata = {
"total_traces": len(traces),
"extracted_at": time.time(),
"source": "checkpoint",
"type": "full_lineage",
}
export_traces_json(traces, output_path, metadata=metadata)
elif format == "jsonl":
# For JSONL, write each trace as a separate line
output_path = Path(output_path)
output_path.parent.mkdir(parents=True, exist_ok=True)
with open(output_path, "w") as f:
for trace in traces:
json.dump(trace, f)
f.write("\n")
else:
raise ValueError(f"Unsupported format: {format}. Use 'json' or 'jsonl'")
logger.info(f"Saved {len(traces)} lineage traces to {output_path}")
return traces
|