Commit 7add95b7 authored by Jan Reimes's avatar Jan Reimes
Browse files

refactor(3gpp-ai): remove AcceleratorConfig - documented no-op, kept for compatibility 

- Removed AcceleratorConfig dataclass (documented as ignored, kept for API compatibility)
- Removed accelerator_config parameter from extract_document_structured()
- Removed accelerator_config parameter from extract_document_text()
- Removed accelerator_config parameter from convert_document_to_markdown()
- Removed accelerator_config parameter from extract_document_structured_from_tdoc()
- Removed all AcceleratorConfig creation and passing in CLI commands
- Removed AcceleratorConfig from __all__ exports in extraction.py
- Cleaned up related comments and del statements
parent 35161e58

packages/3gpp-ai/threegpp_ai/cli.py

+2 −18
Original line number Diff line number Diff line
@@ -90,7 +90,7 @@ from threegpp_ai.models import WorkspaceNotFoundError 
from threegpp_ai.operations.classify import pick_main_document

from threegpp_ai.operations.conversion import OFFICE_FORMATS, convert_to_pdf

from threegpp_ai.operations.convert import convert_document_to_markdown

from threegpp_ai.operations.extraction import AcceleratorConfig, VlmOptions, extract_document_structured

from threegpp_ai.operations.extraction import VlmOptions, extract_document_structured

from threegpp_ai.operations.hybrid_server import (

    DEFAULT_HOST,

    DEFAULT_PORT,
@@ -239,7 +239,6 @@ async def _process_single_item( 
    convert_md: bool = False,

    path_config: PathConfig,

    vlm_options: VlmOptions | None = None,

    accelerator_config: AcceleratorConfig | None = None,

) -> tuple[Any | None, str | None, bool, bool]:

    """Process a single workspace item (checkout + optional PDF conversion + optional markdown extraction).
@@ -253,7 +252,6 @@ async def _process_single_item( 
        convert_md: Whether to extract markdown (implies convert_pdf)

        path_config: PathConfig for file system paths

        vlm_options: Optional VLM features for extraction.

        accelerator_config: Optional accelerator settings for GPU/CPU and threading.



    Returns:

        Tuple of (member, skip_reason, was_converted, was_md_extracted)
@@ -306,7 +304,6 @@ async def _process_single_item( 
                    output_path=None,

                    force=False,

                    vlm_options=vlm_options,

                    accelerator_config=accelerator_config,

                )

            else:

                # Generic extraction (specs, other) - uses file path directly
@@ -317,7 +314,6 @@ async def _process_single_item( 
                        metadata=None,

                        force=False,

                        vlm_options=vlm_options,

                        accelerator_config=accelerator_config,

                    )

            was_md_extracted = True

        except Exception as e:
@@ -409,7 +405,6 @@ async def _process_workspace_members( 
    convert_md: bool = False,

    skip_existing: bool = False,

    vlm_options: VlmOptions | None = None,

    accelerator_config: AcceleratorConfig | None = None,

) -> list[dict[str, Any]]:

    """Process workspace members with optional progress callback.
@@ -421,7 +416,6 @@ async def _process_workspace_members( 
        convert_md: Whether to extract markdown (implies PDF conversion)

        skip_existing: If True, skip extraction for components that already exist.

        vlm_options: Optional VLM features for extraction (enables hybrid mode for figures/tables/equations).

        accelerator_config: Optional accelerator settings for GPU/CPU and threading.



    Returns:

        List of processing results
@@ -490,7 +484,6 @@ async def _process_workspace_members( 
                force=False,

                skip_existing=skip_existing,

                vlm_options=vlm_options,

                accelerator_config=accelerator_config,

            )

            results.append(

                {
@@ -710,7 +703,6 @@ def _checkout_and_convert_items( 
    convert_pdf: bool,

    convert_md: bool,

    vlm_options: VlmOptions | None,

    accelerator_config: AcceleratorConfig,

) -> tuple[list[Any], list[tuple[str, str]], int, int]:

    """Checkout, optionally convert to PDF, and optionally extract markdown for items.
@@ -749,7 +741,6 @@ def _checkout_and_convert_items( 
                    convert_md=convert_md,

                    path_config=manager,

                    vlm_options=vlm_options,

                    accelerator_config=accelerator_config,

                )

                if skip_reason:

                    skipped.append((item, skip_reason))
@@ -803,7 +794,7 @@ def workspace_add_members( 
    kind_normalized = kind.lower().rstrip("s")

    source_kind = SourceKind(kind_normalized) if kind_normalized in {entry.value for entry in SourceKind} else SourceKind.OTHER



    # Build VLM and accelerator options for extraction

    # Build VLM options for extraction

    vlm_options: VlmOptions | None = None

    if vlm:

        # Auto-start hybrid server if not running
@@ -815,8 +806,6 @@ def workspace_add_members( 


        vlm_options = VlmOptions(enable_hybrid=True)



    accelerator_config = AcceleratorConfig(device=device, num_threads=threads, batch_size=batch_size)



    # Phase 1: Resolve items - either directly provided or via database query

    if items is not None:

        resolved_items = items
@@ -855,7 +844,6 @@ def workspace_add_members( 
        convert_pdf=convert_pdf,

        convert_md=convert_md,

        vlm_options=vlm_options,

        accelerator_config=accelerator_config,

    )



    if skipped:
@@ -979,9 +967,6 @@ def workspace_process( 
            raise typer.Exit(1)

        console.print(f"[dim]Using hybrid server at {server_status.url}[/dim]")



    # Build accelerator config from CLI options

    accelerator_config = AcceleratorConfig(device=device, num_threads=threads, batch_size=batch_size)



    with Progress(

        SpinnerColumn(),

        TextColumn("[progress.description]{task.description}"),
@@ -1007,7 +992,6 @@ def workspace_process( 
                convert_md=True,

                skip_existing=skip_existing,

                vlm_options=vlm_options,

                accelerator_config=accelerator_config,

            )

        )

        progress.update(task, completed=len(results), description="[cyan]Processing complete")

packages/3gpp-ai/threegpp_ai/operations/convert.py

+1 −7
Original line number Diff line number Diff line
@@ -24,7 +24,7 @@ from threegpp_ai.operations.conversion import ( 
    ConverterBackend,

    ConverterConfig,

)

from threegpp_ai.operations.extraction import AcceleratorConfig, VlmOptions, extract_document_structured

from threegpp_ai.operations.extraction import VlmOptions, extract_document_structured

from threegpp_ai.operations.extraction_result import StructuredExtractionResult

from threegpp_ai.operations.fetch_tdoc import fetch_tdoc_files

from threegpp_ai.operations.metrics import MetricType, get_metrics_tracker, timed_operation
@@ -168,7 +168,6 @@ def convert_document_to_markdown( 
    force: bool = False,

    converter_config: ConverterConfig | None = None,

    vlm_options: VlmOptions | None = None,

    accelerator_config: AcceleratorConfig | None = None,

) -> str:

    """Convert TDoc to markdown using the unified extraction pipeline.
@@ -188,7 +187,6 @@ def convert_document_to_markdown( 
        force: Force reconversion even if cached

        converter_config: Optional converter configuration (unused, kept for API compatibility)

        vlm_options: Optional VLM features for extraction.

        accelerator_config: Optional accelerator settings for GPU/CPU and threading.



    Returns:

        Markdown content string
@@ -203,7 +201,6 @@ def convert_document_to_markdown( 
        document_id=document_id,

        force=force,

        vlm_options=vlm_options,

        accelerator_config=accelerator_config,

    )



    # Get TDoc metadata for header
@@ -228,7 +225,6 @@ def extract_document_structured_from_tdoc( 
    force: bool = False,

    extract_types: set[str] | None = None,

    vlm_options: VlmOptions | None = None,

    accelerator_config: AcceleratorConfig | None = None,

) -> StructuredExtractionResult:

    """Extract a TDoc into the canonical structured payload.
@@ -241,7 +237,6 @@ def extract_document_structured_from_tdoc( 
        extract_types: Optional set of artifact types to extract/persist.

            If None, extracts all types. Supported types: "tables", "figures", "equations".

        vlm_options: Optional VLM features for extraction.

        accelerator_config: Optional accelerator settings for GPU/CPU and threading.



    Returns:

        Structured extraction result.
@@ -282,7 +277,6 @@ def extract_document_structured_from_tdoc( 
            force=force,

            extract_types=extract_types,

            vlm_options=vlm_options,

            accelerator_config=accelerator_config,

        )

packages/3gpp-ai/threegpp_ai/operations/extraction.py

+0 −27
Original line number Diff line number Diff line
@@ -93,25 +93,6 @@ class VlmOptions: 
    hybrid_fallback: bool = True

    image_output: ImageOutput = ImageOutput.EXTERNAL





@dataclass

class AcceleratorConfig:

    """Accelerator configuration for OpenDataLoader document processing.



    OpenDataLoader is CPU-optimized and doesn't require GPU acceleration.

    This dataclass is kept for API compatibility with the previous docling-based pipeline.



    Attributes:

        device: Compute device (ignored, kept for compatibility).

        num_threads: Thread count for parallel processing.

        batch_size: Batch size for processing (ignored by OpenDataLoader).

    """



    device: str = "auto"

    num_threads: int = 4

    batch_size: int | None = None





# All supported formats (PDF + Office formats + text files)

SUPPORTED_FORMATS = {".pdf", ".txt", ".md"} | OFFICE_FORMATS
@@ -594,7 +575,6 @@ def extract_document_structured( 
    skip_existing: bool = False,

    extract_types: set[str] | None = None,

    vlm_options: VlmOptions | None = None,

    accelerator_config: AcceleratorConfig | None = None,

) -> StructuredExtractionResult:

    """Extract a document into a canonical structured payload.
@@ -620,7 +600,6 @@ def extract_document_structured( 
            If None, extracts all types. Supported types: "tables", "figures", "equations".

        vlm_options: Optional VLM features for extraction. Enables hybrid AI mode

            for complex PDF pages. If None, uses deterministic local extraction.

        accelerator_config: Optional accelerator settings (ignored, kept for compatibility).



    Returns:

        StructuredExtractionResult with content, tables, figures, and equations.
@@ -629,8 +608,6 @@ def extract_document_structured( 
        ExtractionError: If the source file does not exist or extraction fails.

        ConversionError: If Office document conversion fails.

    """

    del accelerator_config  # OpenDataLoader is CPU-optimized, no GPU needed



    if not file_path.exists():

        raise ExtractionError(f"Source file does not exist: {file_path}")
@@ -718,7 +695,6 @@ def extract_document_text( 
    file_path: Path,

    force: bool = False,

    vlm_options: VlmOptions | None = None,

    accelerator_config: AcceleratorConfig | None = None,

) -> str:

    """Extract plain text content from a document using OpenDataLoader.
@@ -729,7 +705,6 @@ def extract_document_text( 
        file_path: Path to the document file.

        force: If True, re-extract even if cached markdown exists.

        vlm_options: Optional VLM features for extraction.

        accelerator_config: Optional accelerator settings (ignored, kept for compatibility).



    Returns:

        Extracted text content as a string.
@@ -744,13 +719,11 @@ def extract_document_text( 
        force=force,

        extract_types=set(),  # No artifacts needed for text-only extraction

        vlm_options=vlm_options,

        accelerator_config=accelerator_config,

    )

    return extraction.content





__all__ = [

    "AcceleratorConfig",

    "VlmOptions",

    "extract_document_structured",

    "extract_document_text",