llm-gguf-tools/helpers/services/orchestrator.py
2025-08-07 18:29:12 +01:00

397 lines
15 KiB
Python

"""Quantisation orchestration service.
High-level orchestration of the complete quantisation workflow from model
acquisition through processing to upload. Manages parallel processing,
status tracking, and cleanup operations for efficient resource utilisation.
"""
from __future__ import annotations
from concurrent.futures import Future, ThreadPoolExecutor
from dataclasses import dataclass, field
from pathlib import Path
from typing import Any
from helpers.config.quantisation_configs import QUANTISATION_CONFIGS, SUPPORTED_QUANTISATION_TYPES
from helpers.logger import logger
from helpers.models.quantisation import (
ModelSource,
QuantisationContext,
QuantisationResult,
QuantisationType,
)
from helpers.services.huggingface import ReadmeGenerator
from helpers.services.llama_cpp import EnvironmentManager, IMatrixGenerator
from helpers.services.quantisation import HuggingFaceUploader, ModelManager, QuantisationEngine
from helpers.utils.tensor_mapping import URLParser
@dataclass(slots=True)
class QuantisationOrchestrator:
"""Orchestrates the complete quantisation workflow.
Uses dataclass with slots for efficient memory usage and dependency injection
for modular service interaction following SOLID principles.
"""
work_dir: Path = field(default_factory=lambda: Path.cwd() / "quantisation_work")
use_imatrix: bool = True
imatrix_base: str = "Q4_K_M"
no_upload: bool = False
# Service dependencies with factory defaults
url_parser: URLParser = field(default_factory=URLParser)
quantisation_engine: QuantisationEngine = field(default_factory=QuantisationEngine)
imatrix_generator: IMatrixGenerator = field(default_factory=IMatrixGenerator)
readme_generator: ReadmeGenerator = field(default_factory=ReadmeGenerator)
uploader: HuggingFaceUploader = field(default_factory=HuggingFaceUploader)
# Computed properties
models_dir: Path = field(init=False)
environment_manager: EnvironmentManager = field(init=False)
model_manager: ModelManager = field(init=False)
def __post_init__(self) -> None:
"""Initialise computed properties after dataclass construction."""
self.models_dir = self.work_dir / "models"
self.environment_manager = EnvironmentManager(self.work_dir)
self.model_manager = ModelManager(self.models_dir, self.environment_manager)
def quantise(self, url: str) -> dict[QuantisationType, QuantisationResult]:
"""Main quantisation workflow orchestrating model processing from URL to upload.
Returns:
dict[QuantisationType, QuantisationResult]: Quantisation results for each type.
"""
logger.info("Starting Bartowski quantisation process...")
# Setup and preparation
model_source, llama_env, f16_model_path, imatrix_path, output_repo = (
self._setup_environment(url)
)
# Create initial repository
self._create_initial_repository(model_source, output_repo)
# Execute all quantisations
results = self._execute_quantisations(
model_source, llama_env, f16_model_path, imatrix_path, output_repo
)
# Cleanup
self._cleanup_files(f16_model_path, model_source)
self._print_completion_summary(model_source, results, output_repo)
return results
def _setup_environment(self, url: str) -> tuple[ModelSource, Any, Path, Path | None, str]:
"""Setup environment and prepare model for quantisation.
Returns:
Tuple of (model_source, llama_env, f16_model_path, imatrix_path, output_repo).
"""
model_source = self.url_parser.parse(url)
self._print_model_info(model_source)
self.models_dir.mkdir(parents=True, exist_ok=True)
llama_env = self.environment_manager.setup()
f16_model_path = self.model_manager.prepare_model(model_source, llama_env)
imatrix_path = None
if self.use_imatrix:
logger.info("Generating importance matrix (imatrix)...")
imatrix_path = self.imatrix_generator.generate_imatrix(
f16_model_path, llama_env, self.models_dir / model_source.model_name
)
output_repo = (
f"{self.uploader.get_username()}/"
f"{model_source.original_author}-{model_source.model_name}-GGUF"
)
return model_source, llama_env, f16_model_path, imatrix_path, output_repo
def _create_initial_repository(self, model_source: ModelSource, output_repo: str) -> None:
"""Create initial repository with planned quantisations."""
logger.info("Creating initial README with planned quantisations...")
planned_results = {
qt: QuantisationResult(quantisation_type=qt, success=False, status="planned")
for qt in SUPPORTED_QUANTISATION_TYPES
}
readme_path = self.readme_generator.generate(
model_source, planned_results, self.models_dir, output_repo
)
if not self.no_upload:
logger.info("Creating repository with planned quantisations...")
self.uploader.upload_readme(output_repo, readme_path)
else:
logger.info("Skipping repository creation (--no-upload specified)")
def _execute_quantisations(
self,
model_source: ModelSource,
llama_env: Any,
f16_model_path: Path,
imatrix_path: Path | None,
output_repo: str,
) -> dict[QuantisationType, QuantisationResult]:
"""Execute all quantisation types with parallel uploads.
Returns:
dict[QuantisationType, QuantisationResult]: Quantisation results for each type.
"""
results: dict[QuantisationType, QuantisationResult] = {}
upload_futures: list[Future[None]] = []
with ThreadPoolExecutor(max_workers=1, thread_name_prefix="uploader") as upload_executor:
for quant_type in SUPPORTED_QUANTISATION_TYPES:
result = self._process_single_quantisation(
quant_type,
model_source,
llama_env,
f16_model_path,
imatrix_path,
output_repo,
results,
upload_executor,
upload_futures,
)
results[quant_type] = result
self._wait_for_uploads(upload_futures)
return results
def _process_single_quantisation(
self,
quant_type: QuantisationType,
model_source: ModelSource,
llama_env: Any,
f16_model_path: Path,
imatrix_path: Path | None,
output_repo: str,
results: dict[QuantisationType, QuantisationResult],
upload_executor: ThreadPoolExecutor,
upload_futures: list,
) -> QuantisationResult:
"""Process a single quantisation type.
Returns:
QuantisationResult: Result of the quantisation attempt.
"""
try:
logger.info(f"Starting {quant_type.value} quantisation...")
config = QUANTISATION_CONFIGS[quant_type]
# Update status to processing
result = QuantisationResult(quantisation_type=quant_type, success=False)
result.status = "processing"
results[quant_type] = result
self._update_readme_status(model_source, results, output_repo)
# Perform quantisation
context = QuantisationContext(
f16_model_path=f16_model_path,
model_source=model_source,
config=config,
llama_env=llama_env,
models_dir=self.models_dir,
imatrix_path=imatrix_path,
base_quant=self.imatrix_base,
)
result = self.quantisation_engine.quantise(context)
self._handle_quantisation_result(
result,
quant_type,
model_source,
results,
output_repo,
upload_executor,
upload_futures,
)
except Exception as e:
return self._handle_quantisation_error(
e, quant_type, model_source, results, output_repo
)
else:
return result
def _handle_quantisation_result(
self,
result: QuantisationResult,
quant_type: QuantisationType,
model_source: ModelSource,
results: dict[QuantisationType, QuantisationResult],
output_repo: str,
upload_executor: ThreadPoolExecutor,
upload_futures: list,
) -> None:
"""Handle successful or failed quantisation result."""
if result.success and result.file_path:
quant_str = getattr(result.quantisation_type, "value", result.quantisation_type)
logger.info(f"Starting parallel upload of {quant_str}...")
upload_future = upload_executor.submit(
self._upload_and_cleanup,
output_repo,
result.file_path,
quant_type,
model_source,
results,
)
upload_futures.append(upload_future)
result.file_path = None # Mark as being uploaded
result.status = "uploading"
else:
result.status = "failed"
self._update_readme_status(model_source, results, output_repo)
def _handle_quantisation_error(
self,
error: Exception,
quant_type: QuantisationType,
model_source: ModelSource,
results: dict[QuantisationType, QuantisationResult],
output_repo: str,
) -> QuantisationResult:
"""Handle quantisation processing error.
Returns:
QuantisationResult: Failed quantisation result with error information.
"""
logger.error(f"Error processing {quant_type.value}: {error}")
result = QuantisationResult(quantisation_type=quant_type, success=False)
result.status = "failed"
result.error_message = str(error)
try:
self._update_readme_status(model_source, results, output_repo)
except Exception as readme_error:
logger.error(f"Failed to update README after error: {readme_error}")
return result
def _update_readme_status(
self,
model_source: ModelSource,
results: dict[QuantisationType, QuantisationResult],
output_repo: str,
) -> None:
"""Update README with current quantisation status."""
if not self.no_upload:
updated_readme_path = self.readme_generator.generate(
model_source, results, self.models_dir, output_repo
)
self.uploader.upload_readme(output_repo, updated_readme_path)
def _wait_for_uploads(self, upload_futures: list) -> None:
"""Wait for all parallel uploads to complete."""
logger.info("Waiting for any remaining uploads to complete...")
for future in upload_futures:
try:
future.result(timeout=300) # 5 minute timeout per upload
except Exception as e:
logger.warning(f"Upload error: {e}")
def _cleanup_files(self, f16_model_path: Path, model_source: ModelSource) -> None:
"""Clean up temporary files after processing."""
if f16_model_path.exists():
logger.info(f"Removing F16 model {f16_model_path.name} to save disk space...")
f16_model_path.unlink()
if not model_source.is_gguf_repo:
self._cleanup_original_model(model_source)
def _cleanup_original_model(self, model_source: ModelSource) -> None:
"""Clean up original safetensors/PyTorch files after successful conversion."""
model_dir = self.models_dir / model_source.model_name
pytorch_files = list(model_dir.glob("pytorch_model*.bin"))
if pytorch_files:
logger.info(f"Removing {len(pytorch_files)} PyTorch model files to save disk space...")
for file in pytorch_files:
file.unlink()
logger.info("Keeping config files, tokeniser, and metadata for reference")
def _upload_and_cleanup(
self,
output_repo: str,
file_path: Path,
quant_type: QuantisationType,
model_source: ModelSource,
results: dict[QuantisationType, QuantisationResult],
) -> None:
"""Upload file and clean up (runs in background thread)."""
try:
logger.info(f"[PARALLEL] Uploading {quant_type}...")
self.uploader.upload_model_file(output_repo, file_path)
logger.info(f"[PARALLEL] Removing {file_path.name} to save disk space...")
file_path.unlink()
results[quant_type].status = "completed"
updated_readme_path = self.readme_generator.generate(
model_source, results, self.models_dir, output_repo
)
self.uploader.upload_readme(output_repo, updated_readme_path)
logger.info(f"[PARALLEL] {quant_type} upload and cleanup complete")
except Exception as e:
logger.error(f"[PARALLEL] Failed to upload {quant_type}: {e}")
results[quant_type].status = "failed"
results[quant_type].error_message = str(e)
updated_readme_path = self.readme_generator.generate(
model_source, results, self.models_dir, output_repo
)
self.uploader.upload_readme(output_repo, updated_readme_path)
raise
def _print_model_info(self, model_source: ModelSource) -> None:
"""Print model information."""
logger.info(f"Source URL: {model_source.url}")
logger.info(f"Source model: {model_source.source_model}")
logger.info(f"Original author: {model_source.original_author}")
logger.info(f"Model name: {model_source.model_name}")
logger.info(f"Your HF username: {self.uploader.get_username()}")
logger.info(f"Working directory: {self.work_dir}")
def _print_completion_summary(
self,
model_source: ModelSource,
results: dict[QuantisationType, QuantisationResult],
output_repo: str,
) -> None:
"""Print completion summary."""
successful_results = [r for r in results.values() if r.success]
if successful_results:
logger.info("Complete! Your quantised models are available at:")
logger.info(f" https://huggingface.co/{output_repo}")
logger.info("Model info:")
logger.info(f" - Source URL: {model_source.url}")
logger.info(f" - Original: {model_source.source_model}")
logger.info(
" - Method: "
f"{'Direct GGUF download' if model_source.is_gguf_repo else 'HF model conversion'}"
)
logger.info(f" - Quantised: {output_repo}")
for result in successful_results:
if result.file_size:
filename = (
f"{model_source.original_author}-{model_source.model_name}-"
f"{result.quantisation_type}.gguf"
)
logger.info(f" - {result.quantisation_type}: {filename} ({result.file_size})")
else:
logger.error(
"All quantisations failed - repository created with documentation "
"but no model files"
)
logger.error(f" Repository: https://huggingface.co/{output_repo}")