Files
qdrant-client/tools/async_client_generator/fastembed_generator.py
George b4ae703d1b Local inference upload collection and upload records (#862)
* wip: add draft implementation of batch processing

* fix: embed dict and list of docs, remove redundant code

* new: regen async, small refactor

* refactor: add docstrings, rename methods

* Upload points local inference (#881)

* new: separate single and plural model embeddings

* fix: fix lazy embed models

* new: add inference object inspections to upload methods

* wip: local inference upload parallel

* new: add local inference to upload points and upload collection, refactor mixin

* fix: remove redundant code

* redundant import

* tests: check is query for query points batch

* refactor: refactor semi ordered map

* tests: add test for local inference with batches with docs and vectors

* tests: check the order of dict processing

* new: distinguish models by options

* fix: fix typing

* fix: fix types

* new: embed batches with different options

* tests: add tests for batch with different options

* fix: ignore ide incorrect type inspection

* tests: wait for points to be inserted

* fix: set threads to 1 in parallel inference

* new: adjust max internal batch size

* fix: fix type hints

* function to get embeddings size (#892)

* function to get embeddings size

* async client

* keep sync

* new: extend embedding size to support image and late interaction models

---------

Co-authored-by: George Panchuk <george.panchuk@qdrant.tech>

* new: add local inference batch size (#894)

---------

Co-authored-by: Andrey Vasnetsov <andrey@vasnetsov.com>
2025-01-29 15:53:24 +01:00

79 lines
2.8 KiB
Python

import inspect
from typing import Optional
from tools.async_client_generator.async_client_base import AsyncQdrantBase
from tools.async_client_generator.base_generator import BaseGenerator
from tools.async_client_generator.transformers import (
ClassDefTransformer,
ImportFromTransformer,
ImportTransformer,
)
from tools.async_client_generator.transformers.fastembed import (
FastembedCallTransformer,
FastembedFunctionDefTransformer,
)
class FastembedGenerator(BaseGenerator):
def __init__(
self,
keep_sync: Optional[list[str]] = None,
class_replace_map: Optional[dict[str, str]] = None,
import_replace_map: Optional[dict[str, str]] = None,
):
super().__init__()
self._async_methods: Optional[list[str]] = None
self.transformers.append(FastembedCallTransformer(async_methods=self.async_methods))
self.transformers.append(ClassDefTransformer(class_replace_map=class_replace_map))
self.transformers.append(ImportTransformer(import_replace_map=import_replace_map))
self.transformers.append(ImportFromTransformer(import_replace_map=import_replace_map))
self.transformers.append(FastembedFunctionDefTransformer(keep_sync=keep_sync))
@property
def async_methods(self) -> list[str]:
if self._async_methods is None:
self._async_methods = self.get_async_methods(AsyncQdrantBase)
return self._async_methods
@staticmethod
def get_async_methods(class_obj: type) -> list[str]:
async_methods = []
for name, method in inspect.getmembers(class_obj):
if inspect.iscoroutinefunction(method):
async_methods.append(name)
return async_methods
if __name__ == "__main__":
from tools.async_client_generator.config import CLIENT_DIR, CODE_DIR
with open(CLIENT_DIR / "qdrant_fastembed.py", "r") as source_file:
code = source_file.read()
generator = FastembedGenerator(
keep_sync=[
"__init__",
"set_model",
"set_sparse_model",
"get_vector_field_name",
"get_sparse_vector_field_name",
"get_embedding_size",
"get_fastembed_vector_params",
"get_fastembed_sparse_vector_params",
"embedding_model_name",
"sparse_embedding_model_name",
],
class_replace_map={
"QdrantBase": "AsyncQdrantBase",
"QdrantFastembedMixin": "AsyncQdrantFastembedMixin",
},
import_replace_map={
"qdrant_client.client_base": "qdrant_client.async_client_base",
"QdrantBase": "AsyncQdrantBase",
},
)
modified_code = generator.generate(code)
with open(CODE_DIR / "async_qdrant_fastembed.py", "w") as target_file:
target_file.write(modified_code)