mirror of
https://github.com/tiennm99/DocsGPT.git
synced 2026-10-03 17:11:24 +00:00
Alex
f6400cd736
Introduces a per-source config contract that makes RAG behavior strategy-dispatched instead of a single hardcoded path. Every source gains a validated JSONB config; an empty/absent config reproduces current behavior byte-for-byte, and the whole path is gated by PER_SOURCE_RETRIEVAL_ENABLED. Foundation: sources.config JSONB column + migration 0022_source_config; SourceConfig/ChunkingConfig/RetrievalConfig pydantic models (strict on write, lenient on read); ChunkerCreator and RetrieverCreator.register registries; config threaded through the upload routes, ingest/remote/connector workers, and reingest. Retrieval: a Dispatcher groups sources by retriever key (all-classic collapses to today's single ClassicRAG under one shared token budget; non-classic retrievers get their own instance), removing the previous single-global-retriever collapse in stream_processor. Per-source chunks, score_threshold (honored for pgvector/mongodb, safely ignored elsewhere), and rephrase_query toggle. New PATCH /api/sources/<id>/config with team-aware (effective_write_owner) authz and a requires_reingest signal. Chunking strategies: recursive, markdown, parent_child (selectable per source; re-ingest to apply). Search exposure: per-source prefetch vs agentic_tool for agentic/research agents. Map-reduce prescreen: optional LLM relevance pre-filter implemented as a composable post-retrieval stage that wraps any retriever. Backend and frontend (shared Retrieval options panel + edit modal) with tests; backend suite and frontend vitest green. Excludes the wiki and GraphRAG flagships.
102 lines
3.5 KiB
Python
102 lines
3.5 KiB
Python
import re
|
|
from typing import List, Tuple
|
|
import logging
|
|
from application.parser.chunking_creator import ChunkerCreator
|
|
from application.parser.schema.base import Document
|
|
from application.utils import get_encoding
|
|
|
|
logger = logging.getLogger(__name__)
|
|
|
|
|
|
class Chunker:
|
|
"""Classic token-window chunker (registered as ``classic_chunk``).
|
|
|
|
Strategy dispatch lives in ``ChunkerCreator``; this class is one
|
|
registered implementation. The ``chunking_strategy`` arg is retained for
|
|
backward-compatible construction and is not used for dispatch here.
|
|
"""
|
|
|
|
def __init__(
|
|
self,
|
|
chunking_strategy: str = "classic_chunk",
|
|
max_tokens: int = 2000,
|
|
min_tokens: int = 150,
|
|
duplicate_headers: bool = False,
|
|
):
|
|
self.chunking_strategy = chunking_strategy
|
|
self.max_tokens = max_tokens
|
|
self.min_tokens = min_tokens
|
|
self.duplicate_headers = duplicate_headers
|
|
self.encoding = get_encoding()
|
|
|
|
def separate_header_and_body(self, text: str) -> Tuple[str, str]:
|
|
header_pattern = r"^(.*?\n){3}"
|
|
match = re.match(header_pattern, text)
|
|
if match:
|
|
header = match.group(0)
|
|
body = text[len(header):]
|
|
else:
|
|
header, body = "", text # No header, treat entire text as body
|
|
return header, body
|
|
|
|
|
|
|
|
def split_document(self, doc: Document) -> List[Document]:
|
|
split_docs = []
|
|
header, body = self.separate_header_and_body(doc.text)
|
|
header_tokens = self.encoding.encode(header) if header else []
|
|
body_tokens = self.encoding.encode(body)
|
|
|
|
current_position = 0
|
|
part_index = 0
|
|
while current_position < len(body_tokens):
|
|
end_position = current_position + self.max_tokens - len(header_tokens)
|
|
chunk_tokens = (header_tokens + body_tokens[current_position:end_position]
|
|
if self.duplicate_headers or part_index == 0 else body_tokens[current_position:end_position])
|
|
chunk_text = self.encoding.decode(chunk_tokens)
|
|
new_doc = Document(
|
|
text=chunk_text,
|
|
doc_id=f"{doc.doc_id}-{part_index}",
|
|
embedding=doc.embedding,
|
|
extra_info={**(doc.extra_info or {}), "token_count": len(chunk_tokens)}
|
|
)
|
|
split_docs.append(new_doc)
|
|
current_position = end_position
|
|
part_index += 1
|
|
header_tokens = []
|
|
return split_docs
|
|
|
|
def classic_chunk(self, documents: List[Document]) -> List[Document]:
|
|
processed_docs = []
|
|
i = 0
|
|
while i < len(documents):
|
|
doc = documents[i]
|
|
tokens = self.encoding.encode(doc.text)
|
|
token_count = len(tokens)
|
|
|
|
if self.min_tokens <= token_count <= self.max_tokens:
|
|
doc.extra_info = doc.extra_info or {}
|
|
doc.extra_info["token_count"] = token_count
|
|
processed_docs.append(doc)
|
|
i += 1
|
|
elif token_count < self.min_tokens:
|
|
|
|
doc.extra_info = doc.extra_info or {}
|
|
doc.extra_info["token_count"] = token_count
|
|
processed_docs.append(doc)
|
|
i += 1
|
|
else:
|
|
# Split large documents
|
|
processed_docs.extend(self.split_document(doc))
|
|
i += 1
|
|
return processed_docs
|
|
|
|
def chunk(
|
|
self,
|
|
documents: List[Document]
|
|
) -> List[Document]:
|
|
return self.classic_chunk(documents)
|
|
|
|
|
|
ChunkerCreator.register("classic_chunk", Chunker)
|