aiqtoolkit 1.1.0a20250516__py3-none-any.whl → 1.1.0a20251020__py3-none-any.whl

aiq/retriever/interface.py

@@ -1,37 +0,0 @@
-# SPDX-FileCopyrightText: Copyright (c) 2025, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
-# SPDX-License-Identifier: Apache-2.0
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-# http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-from abc import ABC
-from abc import abstractmethod
-
-from aiq.retriever.models import RetrieverOutput
-
-
-class AIQRetriever(ABC):
-    """
-    Abstract interface for interacting with data stores.
-
-    A Retriever is resposible for retrieving data from a configured data store.
-
-    Implemntations may integrate with vector stores or other indexing backends that allow for text-based search.
-    """
-
-    @abstractmethod
-    async def search(self, query: str, **kwargs) -> RetrieverOutput:
-        """
-        Retireve max(top_k) items from the data store based on vector similarity search (implementation dependent).
-
-        """
-        raise NotImplementedError

aiq/retriever/milvus/__init__.py

@@ -1,14 +0,0 @@
-# SPDX-FileCopyrightText: Copyright (c) 2025, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
-# SPDX-License-Identifier: Apache-2.0
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-# http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.

aiq/retriever/milvus/register.py

@@ -1,81 +0,0 @@
-# SPDX-FileCopyrightText: Copyright (c) 2025, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
-# SPDX-License-Identifier: Apache-2.0
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-# http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-from pydantic import Field
-from pydantic import HttpUrl
-
-from aiq.builder.builder import Builder
-from aiq.builder.builder import LLMFrameworkEnum
-from aiq.builder.retriever import RetrieverProviderInfo
-from aiq.cli.register_workflow import register_retriever_client
-from aiq.cli.register_workflow import register_retriever_provider
-from aiq.data_models.retriever import RetrieverBaseConfig
-
-
-class MilvusRetrieverConfig(RetrieverBaseConfig, name="milvus_retriever"):
-    """
-    Configuration for a Retriever which pulls data from a Milvus service.
-    """
-    uri: HttpUrl = Field(description="The uri of Milvus service")
-    connection_args: dict = Field(
-        description="Dictionary of arguments used to connect to and authenticate with the Milvus service",
-        default={},
-    )
-    embedding_model: str = Field(description="The name of the embedding model to use for vectorizing the query")
-    collection_name: str | None = Field(description="The name of the milvus collection to search", default=None)
-    content_field: str = Field(description="Name of the primary field to store/retrieve",
-                               default="text",
-                               alias="primary_field")
-    top_k: int | None = Field(gt=0, description="The number of results to return", default=None)
-    output_fields: list[str] | None = Field(
-        default=None,
-        description="A list of fields to return from the datastore. If 'None', all fields but the vector are returned.")
-    search_params: dict = Field(default={"metric_type": "L2"},
-                                description="Search parameters to use when performing vector search")
-    vector_field: str = Field(default="vector", description="Name of the field to compare with the vectorized query")
-    description: str | None = Field(default=None,
-                                    description="If present it will be used as the tool description",
-                                    alias="collection_description")
-
-
-@register_retriever_provider(config_type=MilvusRetrieverConfig)
-async def milvus_retriever(retriever_config: MilvusRetrieverConfig, builder: Builder):
-    yield RetrieverProviderInfo(config=retriever_config,
-                                description="An adapter for a Miluvs data store to use with a Retriever Client")
-
-
-@register_retriever_client(config_type=MilvusRetrieverConfig, wrapper_type=None)
-async def milvus_retriever_client(config: MilvusRetrieverConfig, builder: Builder):
-    from pymilvus import MilvusClient
-
-    from aiq.retriever.milvus.retriever import MilvusRetriever
-
-    embedder = await builder.get_embedder(embedder_name=config.embedding_model, wrapper_type=LLMFrameworkEnum.LANGCHAIN)
-
-    milvus_client = MilvusClient(uri=str(config.uri), **config.connection_args)
-    retriever = MilvusRetriever(
-        client=milvus_client,
-        embedder=embedder,
-        content_field=config.content_field,
-    )
-
-    # Using parameters in the config to set default values which can be overridden during the function call.
-    optional_fields = ["collection_name", "top_k", "output_fields", "search_params", "vector_field"]
-    model_dict = config.model_dump()
-    optional_args = {field: model_dict[field] for field in optional_fields if model_dict[field] is not None}
-
-    retriever.bind(**optional_args)
-
-    yield retriever

aiq/retriever/milvus/retriever.py

@@ -1,228 +0,0 @@
-# SPDX-FileCopyrightText: Copyright (c) 2025, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
-# SPDX-License-Identifier: Apache-2.0
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-# http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-import logging
-from functools import partial
-
-from langchain_core.embeddings import Embeddings
-from pymilvus import MilvusClient
-from pymilvus.client.abstract import Hit
-
-from aiq.retriever.interface import AIQRetriever
-from aiq.retriever.models import AIQDocument
-from aiq.retriever.models import RetrieverError
-from aiq.retriever.models import RetrieverOutput
-
-logger = logging.getLogger(__name__)
-
-
-class CollectionNotFoundError(RetrieverError):
-    pass
-
-
-class MilvusRetriever(AIQRetriever):
-    """
-    Client for retrieving document chunks from a Milvus vectorstore
-    """
-
-    def __init__(
-        self,
-        client: MilvusClient,
-        embedder: Embeddings,
-        content_field: str = "text",
-        use_iterator: bool = False,
-    ) -> None:
-        """
-        Initialize the Milvus Retriever using a preconfigured MilvusClient
-
-        Args:
-           client (MilvusClient): Preinstantiate pymilvus.MilvusClient object.
-        """
-        self._client = client
-        self._embedder = embedder
-
-        if use_iterator and "search_iterator" not in dir(self._client):
-            raise ValueError("This version of the pymilvus.MilvusClient does not support the search iterator.")
-
-        self._search_func = self._search if not use_iterator else self._search_with_iterator
-        self._default_params = None
-        self._bound_params = []
-        self.content_field = content_field
-        logger.info("Mivlus Retriever using %s for search.", self._search_func.__name__)
-
-    def bind(self, **kwargs) -> None:
-        """
-        Bind default values to the search method. Cannot bind the 'query' parameter.
-
-        Args:
-          kwargs (dict): Key value pairs corresponding to the default values of search parameters.
-        """
-        if "query" in kwargs:
-            kwargs = {k: v for k, v in kwargs.items() if k != "query"}
-        self._search_func = partial(self._search_func, **kwargs)
-        self._bound_params = list(kwargs.keys())
-        logger.debug("Binding paramaters for search function: %s", kwargs)
-
-    def get_unbound_params(self) -> list[str]:
-        """
-        Returns a list of unbound parameters which will need to be passed to the search function.
-        """
-        return [param for param in ["query", "collection_name", "top_k", "filters"] if param not in self._bound_params]
-
-    def _validate_collection(self, collection_name: str) -> bool:
-        return collection_name in self._client.list_collections()
-
-    async def search(self, query: str, **kwargs):
-        return await self._search_func(query=query, **kwargs)
-
-    async def _search_with_iterator(self,
-                                    query: str,
-                                    *,
-                                    collection_name: str,
-                                    top_k: int,
-                                    filters: str | None = None,
-                                    output_fields: list[str] | None = None,
-                                    search_params: dict | None = None,
-                                    timeout: float | None = None,
-                                    vector_field_name: str | None = "vector",
-                                    distance_cutoff: float | None = None,
-                                    **kwargs):
-        """
-        Retrieve document chunks from a Milvus vectorstore using a search iterator, allowing for the retrieval of more
-        results.
-        """
-        logger.debug("MilvusRetriever searching query: %s, for collection: %s. Returning max %s results",
-                     query,
-                     collection_name,
-                     top_k)
-
-        if not self._validate_collection(collection_name):
-            raise CollectionNotFoundError(f"Collection: {collection_name} does not exist")
-
-        # If no output fields are specified, return all of them
-        if not output_fields:
-            collection_schema = self._client.describe_collection(collection_name)
-            output_fields = [
-                field["name"] for field in collection_schema.get("fields") if field["name"] != vector_field_name
-            ]
-
-        search_vector = self._embedder.embed_query(query)
-
-        search_iterator = self._client.search_iterator(
-            collection_name=collection_name,
-            data=[search_vector],
-            batch_size=kwargs.get("batch_size", 1000),
-            filter=filters,
-            limit=top_k,
-            output_fields=output_fields,
-            search_params=search_params if search_params else {"metric_type": "L2"},
-            timeout=timeout,
-            anns_field=vector_field_name,
-            round_decimal=kwargs.get("round_decimal", -1),
-            partition_names=kwargs.get("partition_names", None),
-        )
-
-        results = []
-        try:
-            while True:
-                _res = search_iterator.next()
-                res = _res.get_res()
-                if len(_res) == 0:
-                    search_iterator.close()
-                    break
-
-                if distance_cutoff and res[0][-1].distance > distance_cutoff:
-                    for i in range(len(res[0])):
-                        if res[0][i].distance > distance_cutoff:
-                            break
-                        results.append(res[0][i])
-                    break
-                results.extend(res[0])
-
-                return _wrap_milvus_results(results, content_field=self.content_field)
-
-        except Exception as e:
-            logger.exception("Exception when retrieving results from milvus for query %s: %s", query, e)
-            raise RetrieverError(f"Error when retrieving documents from {collection_name} for query '{query}'") from e
-
-    async def _search(self,
-                      query: str,
-                      *,
-                      collection_name: str,
-                      top_k: int,
-                      filters: str | None = None,
-                      output_fields: list[str] | None = None,
-                      search_params: dict | None = None,
-                      timeout: float | None = None,
-                      vector_field_name: str | None = "vector",
-                      **kwargs):
-        """
-        Retrieve document chunks from a Milvus vectorstore
-        """
-        logger.debug("MilvusRetriever searching query: %s, for collection: %s. Returning max %s results",
-                     query,
-                     collection_name,
-                     top_k)
-
-        if not self._validate_collection(collection_name):
-            raise CollectionNotFoundError(f"Collection: {collection_name} does not exist")
-
-        available_fields = [v.get("name") for v in self._client.describe_collection(collection_name).get("fields", {})]
-
-        if self.content_field not in available_fields:
-            raise ValueError(f"The specified content field: {self.content_field} is not part of the schema.")
-
-        if vector_field_name not in available_fields:
-            raise ValueError(f"The specified vector field name: {vector_field_name} is not part of the schema.")
-
-        # If no output fields are specified, return all of them
-        if not output_fields:
-            output_fields = [field for field in available_fields if field != vector_field_name]
-
-        if self.content_field not in output_fields:
-            output_fields.append(self.content_field)
-
-        search_vector = self._embedder.embed_query(query)
-        res = self._client.search(
-            collection_name=collection_name,
-            data=[search_vector],
-            filter=filters,
-            output_fields=output_fields,
-            search_params=search_params if search_params else {"metric_type": "L2"},
-            timeout=timeout,
-            anns_field=vector_field_name,
-            limit=top_k,
-        )
-
-        return _wrap_milvus_results(res[0], content_field=self.content_field)
-
-
-def _wrap_milvus_results(res: list[Hit], content_field: str):
-    return RetrieverOutput(results=[_wrap_milvus_single_results(r, content_field=content_field) for r in res])
-
-
-def _wrap_milvus_single_results(res: Hit | dict, content_field: str) -> AIQDocument:
-    if not isinstance(res, (Hit, dict)):
-        raise ValueError(f"Milvus search returned object of type {type(res)}. Expected 'Hit' or 'dict'.")
-
-    if isinstance(res, Hit):
-        metadata = {k: v for k, v in res.fields.items() if k != content_field}
-        metadata.update({"distance": res.distance})
-        return AIQDocument(page_content=res.fields[content_field], metadata=metadata, document_id=res.id)
-
-    fields = res["entity"]
-    metadata = {k: v for k, v in fields.items() if k != content_field}
-    metadata.update({"distance": res.get("distance")})
-    return AIQDocument(page_content=fields.get(content_field), metadata=metadata, document_id=res["id"])

aiq/retriever/models.py

@@ -1,74 +0,0 @@
-# SPDX-FileCopyrightText: Copyright (c) 2025, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
-# SPDX-License-Identifier: Apache-2.0
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-# http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-from __future__ import annotations
-
-import json
-from typing import Any
-
-from pydantic import BaseModel
-from pydantic import Field
-
-from aiq.utils.type_converter import GlobalTypeConverter
-
-
-class AIQDocument(BaseModel):
-    """
-    Object representing a retrieved document/chunk from a standard AIQ Toolkit Retriever.
-    """
-    page_content: str = Field(description="Primary content of the document to insert or retrieve")
-    metadata: dict[str, Any] = Field(description="Metadata dictionary attached to the AIQDocument")
-    document_id: str | None = Field(description="Unique ID for the document, if supported by the configured datastore",
-                                    default=None)
-
-    @classmethod
-    def from_dict(cls, data: dict[str, Any]) -> AIQDocument:
-        """
-        Deserialize an AIQDocument from a dictionary representation.
-
-        Args:
-            data (dict): A dictionary containing keys
-            'page_content', 'metadata', and optionally 'document_id'.
-
-        Returns:
-            MemoryItem: A reconstructed MemoryItem instance.
-        """
-        return cls(**data)
-
-
-class RetrieverOutput(BaseModel):
-    results: list[AIQDocument] = Field(description="A list of retrieved AIQDocuments")
-
-    def __len__(self):
-        return len(self.results)
-
-    def __str__(self):
-        return json.dumps(self.model_dump())
-
-
-class RetrieverError(Exception):
-    pass
-
-
-def retriever_output_to_dict(obj: RetrieverOutput) -> dict:
-    return obj.model_dump()
-
-
-def retriever_output_to_str(obj: RetrieverOutput) -> str:
-    return str(obj)
-
-
-GlobalTypeConverter.register_converter(retriever_output_to_dict)
-GlobalTypeConverter.register_converter(retriever_output_to_str)

aiq/retriever/nemo_retriever/__init__.py

@@ -1,14 +0,0 @@
-# SPDX-FileCopyrightText: Copyright (c) 2025, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
-# SPDX-License-Identifier: Apache-2.0
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-# http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.

aiq/retriever/nemo_retriever/register.py

@@ -1,60 +0,0 @@
-# SPDX-FileCopyrightText: Copyright (c) 2025, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
-# SPDX-License-Identifier: Apache-2.0
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-# http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-from pydantic import Field
-from pydantic import HttpUrl
-
-from aiq.builder.builder import Builder
-from aiq.builder.retriever import RetrieverProviderInfo
-from aiq.cli.register_workflow import register_retriever_client
-from aiq.cli.register_workflow import register_retriever_provider
-from aiq.data_models.retriever import RetrieverBaseConfig
-
-
-class NemoRetrieverConfig(RetrieverBaseConfig, name="nemo_retriever"):
-    """
-    Configuration for a Retriever which pulls data from a Nemo Retriever service.
-    """
-    uri: HttpUrl = Field(description="The uri of the Nemo Retriever service.")
-    collection_name: str | None = Field(description="The name of the collection to search", default=None)
-    top_k: int | None = Field(description="The number of results to return", gt=0, le=50, default=None)
-    output_fields: list[str] | None = Field(
-        default=None,
-        description="A list of fields to return from the datastore. If 'None', all fields but the vector are returned.")
-    timeout: int = Field(default=60, description="Maximum time to wait for results to be returned from the service.")
-    nvidia_api_key: str | None = Field(
-        description="API key used to authenticate with the service. If 'None', will use ENV Variable 'NVIDIA_API_KEY'",
-        default=None,
-    )
-
-
-@register_retriever_provider(config_type=NemoRetrieverConfig)
-async def nemo_retriever(retriever_config: NemoRetrieverConfig, builder: Builder):
-    yield RetrieverProviderInfo(config=retriever_config,
-                                description="An adapter for a  Nemo data store for use with a Retriever Client")
-
-
-@register_retriever_client(config_type=NemoRetrieverConfig, wrapper_type=None)
-async def nemo_retriever_client(config: NemoRetrieverConfig, builder: Builder):
-    from aiq.retriever.nemo_retriever.retriever import NemoRetriever
-
-    retriever = NemoRetriever(**config.model_dump(exclude={"type", "top_k", "collection_name"}))
-    optional_fields = ["collection_name", "top_k", "output_fields"]
-    model_dict = config.model_dump()
-    optional_args = {field: model_dict[field] for field in optional_fields if model_dict[field] is not None}
-
-    retriever.bind(**optional_args)
-
-    yield retriever