File size: 11,818 Bytes
af83196 | 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 | """
Evox context builder: co-evolves search algorithms and solutions.
"""
import asyncio
import concurrent.futures
import hashlib
import logging
from pathlib import Path
from typing import Any, Dict, List, Union
from skydiscover.config import Config
from skydiscover.context_builder.default.builder import DefaultContextBuilder
from skydiscover.context_builder.evox import formatters as fmt
from skydiscover.context_builder.utils import TemplateManager, prog_attr
from skydiscover.llm.llm_pool import LLMPool
from skydiscover.search.base_database import Program
def run_async_safely(coro):
"""Run an async coroutine, handling nested event loops."""
try:
asyncio.get_running_loop()
except RuntimeError:
return asyncio.run(coro)
else:
with concurrent.futures.ThreadPoolExecutor(max_workers=1) as executor:
future = executor.submit(asyncio.run, coro)
return future.result()
logger = logging.getLogger(__name__)
_DEFAULT_TEMPLATES_DIR = str(Path(__file__).parent.parent / "default" / "templates")
_EVOX_TEMPLATES_DIR = str(Path(__file__).parent / "templates")
class EvoxContextBuilder(DefaultContextBuilder):
"""Context builder for search algorithm discovery."""
def __init__(
self,
config: Config,
use_llm_stats_insight: bool = True,
):
super().__init__(config)
self.use_llm_stats_insight = use_llm_stats_insight
self.template_manager = TemplateManager(_DEFAULT_TEMPLATES_DIR, _EVOX_TEMPLATES_DIR)
summary_llm_config = config.llm.guide_models
self.summary_llm: LLMPool = LLMPool(summary_llm_config)
if summary_llm_config:
logger.info(
f"Initialized guide LLM inside EvoxContextBuilder: {summary_llm_config[0].name}"
)
self._problem_context_summary_cache: Dict[str, str] = {}
evox_search_sys_prompt_path = (
Path(__file__).parent.parent.parent
/ "search"
/ "evox"
/ "config"
/ "evox_search_sys_prompt.txt"
)
with open(evox_search_sys_prompt_path, "r") as f:
self.relevant_task_description_message = f.read()
batch_prompt = self.template_manager.get_template("batch_summary_prompt")
self._batch_sections = self._parse_template_sections(batch_prompt)
@staticmethod
def _parse_template_sections(text: str) -> Dict[str, str]:
"""Parse a template with ===SECTION=== markers into a dict of section_name -> content."""
sections: Dict[str, str] = {}
current_section = None
current_lines: List[str] = []
for line in text.splitlines():
stripped = line.strip()
if stripped.startswith("===") and stripped.endswith("===") and len(stripped) > 6:
if current_section is not None:
sections[current_section] = "\n".join(current_lines).strip()
current_section = stripped[3:-3]
current_lines = []
else:
current_lines.append(line)
if current_section is not None:
sections[current_section] = "\n".join(current_lines).strip()
return sections
async def _generate_stats_insight_async(self, stats_text: str) -> str:
"""Generate stats insight via LLM."""
if not stats_text:
return ""
user_content = f"Population Statistics:\n\n{stats_text}"
result = await self.summary_llm.generate(
system_message=self.template_manager.get_template("stats_insight_system_message"),
messages=[{"role": "user", "content": user_content}],
)
return result.text
async def _generate_problem_context_summary_async(
self, problem_description: str, evaluator_context: str
) -> str:
"""Generate problem context summary via LLM (cached)."""
cache_key = hashlib.sha256(
f"{problem_description}|||{evaluator_context}".encode("utf-8")
).hexdigest()
if cache_key in self._problem_context_summary_cache:
return self._problem_context_summary_cache[cache_key]
problem_context_input = self.template_manager.get_template("problem_template").format(
problem_description=problem_description,
evaluator_context=evaluator_context,
)
result = await self.summary_llm.generate(
system_message=self.template_manager.get_template(
"problem_context_summary_system_message"
),
messages=[{"role": "user", "content": problem_context_input}],
)
self._problem_context_summary_cache[cache_key] = result.text
return result.text
async def _generate_batch_summaries_async(self, batch_user_message: str) -> str:
"""Generate batch summaries via LLM."""
result = await self.summary_llm.generate(
system_message=self._batch_sections["SYSTEM"],
messages=[{"role": "user", "content": batch_user_message}],
)
return result.text
def build_prompt(
self,
current_program: Union[Program, Dict[str, Program]],
context: Dict[str, Any] = None,
**kwargs: Any,
) -> Dict[str, str]:
"""Build prompt for search algorithm discovery.
Args:
current_program: Program or {label: Program} to evolve from.
context: optional dict with program_metrics, other_context_programs,
previous_programs, db_stats, search_stats, etc.
Returns:
{"system": str, "user": str} ready for LLM.generate().
"""
context = context or {}
program_metrics = context.get("program_metrics", {})
other_context_programs = context.get("other_context_programs", {})
previous_programs = context.get("previous_programs", [])
language = self.config.language or "python"
user_template_key = self.user_template_override or "search_evolution_user_message"
user_template = self.template_manager.get_template(user_template_key)
system_message = self._get_system_message()
search_window_context = fmt.format_search_window_context(context)
db_stats = context.get("db_stats", {})
actual_program = (
list(current_program.values())[0]
if isinstance(current_program, dict) and current_program
else current_program
)
horizon = 0
if actual_program:
metrics = prog_attr(actual_program, "metrics", {})
horizon = int(metrics.get("search_window_horizon") or 0)
if horizon > 0 and db_stats:
db_stats = fmt.filter_db_stats_by_horizon(db_stats, horizon)
search_stats = context.get("search_stats") or {}
stats_insight_data = None
population_state = ""
if self.use_llm_stats_insight and db_stats:
stats_text = fmt.format_population_state(db_stats)
if stats_text:
stats_insight_data = stats_text
else:
population_state = fmt.format_population_state(db_stats) if db_stats else ""
problem_description = fmt.format_problem_description(
search_stats.get("problem_description")
)
evaluator_context = fmt.format_evaluator_context(search_stats.get("evaluator_context"))
problem_context_data = self.template_manager.get_template("problem_template").format(
problem_description=problem_description,
evaluator_context=evaluator_context,
)
all_programs_data = fmt.prepare_search_algorithms_data(other_context_programs)
batch_summary_data = None
if all_programs_data:
per_program_tpl = self._batch_sections["PER_PROGRAM"]
combined_content_parts = []
for prog in all_programs_data:
part = (
f"=== PROGRAM {prog['program_num']} (score={prog['combined_score']:.4f}, improvement={prog['improvement']:.4f}) ===\n"
+ per_program_tpl.format(
task_description=self.relevant_task_description_message,
solution=prog["solution"],
db_stats_text=prog["db_stats_text"],
)
)
combined_content_parts.append(part)
batch_summary_data = self._batch_sections["INSTRUCTIONS"].format(
num_programs=len(all_programs_data),
combined_content="\n".join(combined_content_parts),
)
async def gather_llm_calls():
tasks = []
if stats_insight_data:
tasks.append(
("stats_insight", self._generate_stats_insight_async(stats_insight_data))
)
has_meaningful_data = (
problem_description
and problem_description.strip()
and evaluator_context
and evaluator_context.strip()
and not (
problem_description.startswith("(No ") and evaluator_context.startswith("(No ")
)
)
if has_meaningful_data:
tasks.append(
(
"problem_context",
self._generate_problem_context_summary_async(
problem_description, evaluator_context
),
)
)
if batch_summary_data:
tasks.append(
("batch_summaries", self._generate_batch_summaries_async(batch_summary_data))
)
if not tasks:
return {}
results = await asyncio.gather(*[task for _, task in tasks], return_exceptions=True)
result_dict = {}
for (name, _), result in zip(tasks, results):
if isinstance(result, Exception):
result_dict[name] = ""
else:
result_dict[name] = result
return result_dict
llm_results = run_async_safely(gather_llm_calls())
if self.use_llm_stats_insight:
population_state = llm_results.get("stats_insight", population_state)
problem_context_summary = llm_results.get("problem_context", "")
problem_template = (
problem_context_summary.strip()
if problem_context_summary
else (problem_context_data or "")
)
summaries_by_num = fmt.parse_batch_summaries(
llm_results.get("batch_summaries", ""), all_programs_data
)
simplification_threshold = self.context_config.suggest_simplification_after_chars
improvement_areas = fmt.identify_search_improvement_areas(
actual_program, program_metrics, previous_programs, simplification_threshold
)
other_context_program_section = fmt.format_search_algorithms(
other_context_programs, language, summaries_by_num=summaries_by_num
)
current_program_section = fmt.format_current_program(
current_program,
language,
improvement_areas=improvement_areas,
)
user_message = user_template.format(
current_program=current_program_section,
other_context_programs=other_context_program_section,
language=language,
search_window_context=search_window_context,
population_state=population_state,
problem_template=problem_template,
**kwargs,
)
return {
"system": system_message,
"user": user_message,
}
|