mirror of
https://github.com/trustgraph-ai/trustgraph.git
synced 2026-04-25 00:16:23 +02:00
24bbe94136
- Schema - ChunkEmbeddings now uses chunk_id: str instead of chunk: bytes
- Schema - DocumentEmbeddingsResponse now returns chunk_ids: list[str]
instead of chunks
- Translators - Updated to serialize/deserialize chunk_id
- Clients - DocumentEmbeddingsClient.query() returns chunk_ids
- SDK/API - flow.py, socket_client.py, bulk_client.py updated
- Document embeddings service - Stores chunk_id (document ID) instead
of chunk text
- Storage writers - Qdrant, Milvus, Pinecone store chunk_id in payload
- Query services - Return chunk_id from vector store searches
- Gateway dispatchers - Serialize chunk_id in API responses
- Document RAG - Added librarian client to fetch chunk content from
Garage using chunk_ids
- CLI tools - Updated all three tools:
- invoke_document_embeddings.py - displays chunk_ids, removed
max_chunk_length
- save_doc_embeds.py - exports chunk_id
- load_doc_embeds.py - imports chunk_id
101 lines
2.3 KiB
Python
Executable file
101 lines
2.3 KiB
Python
Executable file
|
|
"""
|
|
Document embeddings, calls the embeddings service to get embeddings for a
|
|
chunk of text. Input is chunk of text plus metadata.
|
|
Output is chunk plus embedding.
|
|
"""
|
|
|
|
from ... schema import Chunk, ChunkEmbeddings, DocumentEmbeddings
|
|
from ... schema import EmbeddingsRequest, EmbeddingsResponse
|
|
|
|
from ... base import FlowProcessor, RequestResponseSpec, ConsumerSpec
|
|
from ... base import ProducerSpec
|
|
|
|
import logging
|
|
|
|
logger = logging.getLogger(__name__)
|
|
|
|
default_ident = "document-embeddings"
|
|
|
|
class Processor(FlowProcessor):
|
|
|
|
def __init__(self, **params):
|
|
|
|
id = params.get("id")
|
|
|
|
super(Processor, self).__init__(
|
|
**params | {
|
|
"id": id,
|
|
}
|
|
)
|
|
|
|
self.register_specification(
|
|
ConsumerSpec(
|
|
name = "input",
|
|
schema = Chunk,
|
|
handler = self.on_message,
|
|
)
|
|
)
|
|
|
|
self.register_specification(
|
|
RequestResponseSpec(
|
|
request_name = "embeddings-request",
|
|
request_schema = EmbeddingsRequest,
|
|
response_name = "embeddings-response",
|
|
response_schema = EmbeddingsResponse,
|
|
)
|
|
)
|
|
|
|
self.register_specification(
|
|
ProducerSpec(
|
|
name = "output",
|
|
schema = DocumentEmbeddings
|
|
)
|
|
)
|
|
|
|
async def on_message(self, msg, consumer, flow):
|
|
|
|
v = msg.value()
|
|
logger.info(f"Indexing {v.metadata.id}...")
|
|
|
|
try:
|
|
|
|
resp = await flow("embeddings-request").request(
|
|
EmbeddingsRequest(
|
|
text = v.chunk
|
|
)
|
|
)
|
|
|
|
vectors = resp.vectors
|
|
|
|
embeds = [
|
|
ChunkEmbeddings(
|
|
chunk_id=v.document_id,
|
|
vectors=vectors,
|
|
)
|
|
]
|
|
|
|
r = DocumentEmbeddings(
|
|
metadata=v.metadata,
|
|
chunks=embeds,
|
|
)
|
|
|
|
await flow("output").send(r)
|
|
|
|
except Exception as e:
|
|
logger.error("Exception occurred", exc_info=True)
|
|
|
|
# Retry
|
|
raise e
|
|
|
|
logger.info("Done.")
|
|
|
|
@staticmethod
|
|
def add_args(parser):
|
|
|
|
FlowProcessor.add_args(parser)
|
|
|
|
def run():
|
|
|
|
Processor.launch(default_ident, __doc__)
|
|
|