File size: 19,103 Bytes
03f0acb | 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 | """Agentic text-to-image prompt upsampling orchestration."""
from __future__ import annotations
import json
import traceback
from concurrent.futures import ThreadPoolExecutor, as_completed
from dataclasses import dataclass
from pathlib import Path
from typing import Any, Protocol
from agentic_upsampling.clients import GenerationOutput
from agentic_upsampling.constants import DEFAULT_JPEG_QUALITY, DEFAULT_MAX_ITERATIONS, DEFAULT_SAMPLES_PER_ITERATION
from agentic_upsampling.data import PromptItem, prompt_dir_name
from agentic_upsampling.io_utils import read_json, write_json_atomic
from agentic_upsampling.rubric import candidate_sort_key
class RewriterLike(Protocol):
def initial_prompt(self, item: PromptItem) -> dict[str, Any]:
"""Create an initial prompt."""
def rewrite_prompt_pair(
self,
item: PromptItem,
previous_prompt: dict[str, Any],
previous_negative_prompt: str,
previous_analysis: dict[str, Any],
history: list[dict[str, Any]],
) -> tuple[dict[str, Any], str]:
"""Jointly rewrite a positive prompt and negative prompt."""
class GeneratorLike(Protocol):
def generate(
self,
*,
prompt_json: dict[str, Any],
prompt_id: str,
output_dir: Path,
seed: int | None = None,
negative_prompt: str = "",
jpeg_quality: int = DEFAULT_JPEG_QUALITY,
) -> GenerationOutput:
"""Generate one image."""
class JudgeLike(Protocol):
def score_image(
self,
*,
item: PromptItem,
image_path: Path,
) -> dict[str, Any]:
"""Score one image."""
@dataclass(frozen=True, slots=True)
class RunnerConfig:
"""Runtime settings for the agentic loop."""
output_dir: Path
max_iterations: int = DEFAULT_MAX_ITERATIONS
samples_per_iteration: int = DEFAULT_SAMPLES_PER_ITERATION
overwrite: bool = False
seed_base: int | None = None
jpeg_quality: int = DEFAULT_JPEG_QUALITY
initial_negative_prompt: str = ""
early_stop: bool = True
verbose: bool = True
def __post_init__(self) -> None:
if self.max_iterations < 1:
raise ValueError("max_iterations must be >= 1.")
if self.samples_per_iteration < 1:
raise ValueError("samples_per_iteration must be >= 1.")
@dataclass(frozen=True, slots=True)
class IterationPrompt:
"""Positive and negative prompts prepared for one iteration."""
prompt_json: dict[str, Any]
negative_prompt: str
class AgenticUpsamplerRunner:
"""Run the iterative prompt rewrite, generate, and judge loop."""
rewriter: RewriterLike
generator: GeneratorLike
judge: JudgeLike
config: RunnerConfig
def __init__(
self,
*,
rewriter: RewriterLike,
generator: GeneratorLike,
judge: JudgeLike,
config: RunnerConfig,
) -> None:
self.rewriter = rewriter
self.generator = generator
self.judge = judge
self.config = config
def run_item(self, item: PromptItem) -> dict[str, Any]:
"""Run all iterations for one prompt item and persist the best candidate."""
item_dir = self.config.output_dir / prompt_dir_name(item)
item_dir.mkdir(parents=True, exist_ok=True)
(item_dir / "failure.json").unlink(missing_ok=True)
(item_dir / "incomplete.json").unlink(missing_ok=True)
self._log(f"[prompt {item.prompt_id}] start")
candidates: list[dict[str, Any]] = []
previous_prompt: dict[str, Any] | None = None
previous_analysis: dict[str, Any] | None = None
previous_negative_prompt = self.config.initial_negative_prompt.strip()
incomplete_error: dict[str, Any] | None = None
for iteration in range(self.config.max_iterations):
iteration_dir = item_dir / f"iter_{iteration:02d}"
candidate = None if self.config.overwrite else self._load_iteration(iteration_dir, iteration)
if candidate is None:
try:
candidate = self._run_iteration(
item,
iteration_dir,
iteration,
previous_prompt,
previous_analysis,
previous_negative_prompt,
candidates,
)
except Exception as exc:
if not candidates:
raise
incomplete_error = {
"iteration": iteration,
"error": repr(exc),
"traceback": traceback.format_exc(),
}
write_json_atomic(item_dir / "incomplete.json", incomplete_error)
self._log(f"[prompt {item.prompt_id}] incomplete at iter={iteration}: {exc!r}")
break
candidates.append(candidate)
previous_prompt = candidate["prompt_json"]
previous_analysis = candidate["analysis"]
previous_negative_prompt = str(candidate.get("negative_prompt") or "")
if self.config.early_stop and bool(candidate["analysis"].get("threshold_cleared")):
self._log(f"[prompt {item.prompt_id}] early stop at iter={iteration}")
break
return self.finalize_item(item, candidates, incomplete_error=incomplete_error)
def run_item_safely(self, item: PromptItem) -> dict[str, Any]:
"""Run one item and convert failures into structured records."""
try:
return self.run_item(item)
except Exception as exc:
self._log(f"[prompt {item.prompt_id}] failed: {exc!r}")
failure = {
"prompt_id": item.prompt_id,
"prompt": item.prompt,
"error": repr(exc),
"traceback": traceback.format_exc(),
}
failure_path = self.config.output_dir / prompt_dir_name(item) / "failure.json"
write_json_atomic(failure_path, failure)
return {"prompt_id": item.prompt_id, "error": repr(exc), "failure_path": str(failure_path)}
def _run_iteration(
self,
item: PromptItem,
iteration_dir: Path,
iteration: int,
previous_prompt: dict[str, Any] | None,
previous_analysis: dict[str, Any] | None,
previous_negative_prompt: str,
candidates: list[dict[str, Any]],
) -> dict[str, Any]:
prepared = self.prepare_iteration_prompt(
item,
iteration_dir,
iteration,
previous_prompt,
previous_analysis,
previous_negative_prompt,
candidates,
)
sample_candidates, sample_errors = self._run_iteration_samples(
item,
iteration_dir,
iteration,
prepared.prompt_json,
prepared.negative_prompt,
)
return self.finalize_iteration(item, iteration_dir, iteration, sample_candidates, sample_errors)
def _run_iteration_samples(
self,
item: PromptItem,
iteration_dir: Path,
iteration: int,
prompt_json: dict[str, Any],
negative_prompt: str,
) -> tuple[list[dict[str, Any]], list[dict[str, Any]]]:
"""Generate seed samples concurrently, then judge successful images in sample order."""
generation_outputs: dict[int, GenerationOutput] = {}
sample_errors: list[dict[str, Any]] = []
with ThreadPoolExecutor(max_workers=self.config.samples_per_iteration) as executor:
future_to_sample_index = {
executor.submit(
self.run_generation_sample,
item,
iteration_dir,
sample_index,
prompt_json,
negative_prompt,
): sample_index
for sample_index in range(self.config.samples_per_iteration)
}
for future in as_completed(future_to_sample_index):
sample_index = future_to_sample_index[future]
try:
generation_outputs[sample_index] = future.result()
except Exception as exc:
sample_errors.append(self._record_sample_error(item, iteration_dir, iteration, sample_index, exc))
sample_candidates: list[dict[str, Any]] = []
for sample_index in range(self.config.samples_per_iteration):
generation = generation_outputs.get(sample_index)
if generation is None:
continue
try:
sample_candidates.append(
self.judge_iteration_sample(
item,
iteration_dir,
iteration,
sample_index,
prompt_json,
negative_prompt,
generation,
)
)
except Exception as exc:
sample_errors.append(self._record_sample_error(item, iteration_dir, iteration, sample_index, exc))
return sample_candidates, sample_errors
def _record_sample_error(
self,
item: PromptItem,
iteration_dir: Path,
iteration: int,
sample_index: int,
exc: Exception,
) -> dict[str, Any]:
"""Persist one per-sample failure record."""
error = {"sample_index": sample_index, "error": repr(exc), "traceback": traceback.format_exc()}
write_json_atomic(self._sample_dir(iteration_dir, sample_index) / "failure.json", error)
self._log(f"[prompt {item.prompt_id}] iter={iteration} sample={sample_index} failed: {exc!r}")
return error
def prepare_iteration_prompt(
self,
item: PromptItem,
iteration_dir: Path,
iteration: int,
previous_prompt: dict[str, Any] | None,
previous_analysis: dict[str, Any] | None,
previous_negative_prompt: str,
candidates: list[dict[str, Any]],
) -> IterationPrompt:
"""Prepare and persist the positive/negative prompt pair for one iteration."""
iteration_dir.mkdir(parents=True, exist_ok=True)
self._log(f"[prompt {item.prompt_id}] iter={iteration} start")
if iteration == 0 or previous_prompt is None or previous_analysis is None:
prompt_json = self.rewriter.initial_prompt(item)
negative_prompt = self.config.initial_negative_prompt.strip()
else:
prompt_json, negative_prompt = self.rewriter.rewrite_prompt_pair(
item,
previous_prompt,
previous_negative_prompt,
previous_analysis,
candidates,
)
negative_prompt = negative_prompt.strip()
write_json_atomic(iteration_dir / "prompt.json", prompt_json)
write_json_atomic(iteration_dir / "negative_prompt.json", {"negative_prompt": negative_prompt})
return IterationPrompt(prompt_json=prompt_json, negative_prompt=negative_prompt)
def _run_iteration_sample(
self,
item: PromptItem,
iteration_dir: Path,
iteration: int,
sample_index: int,
prompt_json: dict[str, Any],
negative_prompt: str,
) -> dict[str, Any]:
generation = self.run_generation_sample(item, iteration_dir, sample_index, prompt_json, negative_prompt)
return self.judge_iteration_sample(
item,
iteration_dir,
iteration,
sample_index,
prompt_json,
negative_prompt,
generation,
)
def run_generation_sample(
self,
item: PromptItem,
iteration_dir: Path,
sample_index: int,
prompt_json: dict[str, Any],
negative_prompt: str,
) -> GenerationOutput:
"""Generate one sample image for an iteration."""
sample_dir = self._sample_dir(iteration_dir, sample_index)
sample_dir.mkdir(parents=True, exist_ok=True)
self._log(f"[prompt {item.prompt_id}] sample={sample_index} generate")
return self.generator.generate(
prompt_json=prompt_json,
prompt_id=item.prompt_id,
output_dir=sample_dir,
seed=self._sample_seed(sample_index),
negative_prompt=negative_prompt,
jpeg_quality=self.config.jpeg_quality,
)
def judge_iteration_sample(
self,
item: PromptItem,
iteration_dir: Path,
iteration: int,
sample_index: int,
prompt_json: dict[str, Any],
negative_prompt: str,
generation: GenerationOutput,
) -> dict[str, Any]:
"""Judge one generated sample and persist its candidate metadata."""
sample_dir = self._sample_dir(iteration_dir, sample_index)
analysis = self.judge.score_image(item=item, image_path=generation.image_path)
self._log(f"[prompt {item.prompt_id}] iter={iteration} sample={sample_index} score={analysis.get('overall_score')}")
analysis_path = sample_dir / "analysis.json"
write_json_atomic(analysis_path, analysis)
candidate = {
"prompt_id": item.prompt_id,
"iteration": iteration,
"sample_index": sample_index,
"prompt_path": str(iteration_dir / "prompt.json"),
"image_path": str(generation.image_path),
"analysis_path": str(analysis_path),
"generation_meta_path": str(generation.meta_path),
"negative_prompt_path": str(iteration_dir / "negative_prompt.json"),
"negative_prompt": negative_prompt,
"prompt_json": prompt_json,
"analysis": analysis,
}
write_json_atomic(sample_dir / "meta.json", candidate)
return candidate
def finalize_iteration(
self,
item: PromptItem,
iteration_dir: Path,
iteration: int,
sample_candidates: list[dict[str, Any]],
sample_errors: list[dict[str, Any]],
) -> dict[str, Any]:
"""Select and persist the best sample candidate for one iteration."""
if not sample_candidates:
raise RuntimeError(f"All {self.config.samples_per_iteration} samples failed for iteration {iteration}.")
write_json_atomic(iteration_dir / "samples.json", sample_candidates)
candidate = dict(max(sample_candidates, key=candidate_sort_key))
candidate["samples"] = sample_candidates
candidate["sample_count"] = len(sample_candidates)
candidate["selected_sample_index"] = candidate["sample_index"]
if sample_errors:
candidate["sample_errors"] = sample_errors
write_json_atomic(iteration_dir / "sample_failures.json", sample_errors)
write_json_atomic(iteration_dir / "meta.json", candidate)
self._log(
f"[prompt {item.prompt_id}] iter={iteration} best_sample={candidate['selected_sample_index']} "
f"score={candidate['analysis'].get('overall_score')} samples={len(sample_candidates)}"
)
return candidate
def finalize_item(
self,
item: PromptItem,
candidates: list[dict[str, Any]],
*,
incomplete_error: dict[str, Any] | None = None,
) -> dict[str, Any]:
"""Persist and return the best candidate summary for a completed or incomplete item."""
if not candidates:
raise RuntimeError(f"No candidates produced for prompt {item.prompt_id}.")
item_dir = self.config.output_dir / prompt_dir_name(item)
best = max(candidates, key=candidate_sort_key)
summary = {
"prompt_id": item.prompt_id,
"prompt": item.prompt,
"best_iteration": best["iteration"],
"best_score": best["analysis"].get("overall_score"),
"threshold_cleared_any": any(bool(candidate["analysis"].get("threshold_cleared")) for candidate in candidates),
"best": best,
"iterations": candidates,
}
if incomplete_error is not None:
summary["incomplete_error"] = incomplete_error
write_json_atomic(item_dir / "best.json", summary)
self._log(f"[prompt {item.prompt_id}] done best_iter={summary['best_iteration']} best_score={summary['best_score']}")
return summary
def _log(self, message: str) -> None:
if self.config.verbose:
print(message, flush=True)
def _sample_seed(self, sample_index: int) -> int | None:
if self.config.seed_base is None:
return None
return self.config.seed_base + sample_index
def _sample_dir(self, iteration_dir: Path, sample_index: int) -> Path:
if self.config.samples_per_iteration == 1:
return iteration_dir
return iteration_dir / f"sample_{sample_index:02d}"
@staticmethod
def _load_iteration(iteration_dir: Path, iteration: int) -> dict[str, Any] | None:
meta_path = iteration_dir / "meta.json"
prompt_path = iteration_dir / "prompt.json"
if not (meta_path.exists() and prompt_path.exists()):
return None
meta = read_json(meta_path)
analysis_path = Path(str(meta.get("analysis_path") or iteration_dir / "analysis.json"))
image_path = Path(str(meta.get("image_path") or iteration_dir / "image.jpg"))
if not (analysis_path.exists() and image_path.exists()):
return None
meta["iteration"] = iteration
meta["prompt_json"] = read_json(prompt_path)
meta["analysis"] = read_json(analysis_path)
negative_prompt_path = iteration_dir / "negative_prompt.json"
if "negative_prompt" not in meta and negative_prompt_path.exists():
negative_prompt_data = read_json(negative_prompt_path)
meta["negative_prompt"] = str(negative_prompt_data.get("negative_prompt") or "")
meta["negative_prompt_path"] = str(negative_prompt_path)
meta.setdefault("negative_prompt", "")
samples_path = iteration_dir / "samples.json"
if samples_path.exists():
samples = json.loads(samples_path.read_text(encoding="utf-8"))
if isinstance(samples, list):
meta["samples"] = samples
meta["sample_count"] = len(samples)
return meta
def write_run_manifest(output_dir: Path, results: list[dict[str, Any]]) -> None:
"""Write compact run-level manifest files."""
manifest_path = output_dir / "manifest.jsonl"
failures_path = output_dir / "failures.jsonl"
manifest_path.unlink(missing_ok=True)
failures_path.unlink(missing_ok=True)
for result in results:
target = failures_path if result.get("error") else manifest_path
with target.open("a", encoding="utf-8") as f:
f.write(json.dumps(result, ensure_ascii=True, separators=(",", ":")) + "\n")
|