Skip to content

armgit / camel

Public Imported from GitHub
0

🐫 CAMEL: The first and the best multi-agent framework. Finding the Scaling Law of Agents. https://www.camel-ai.org

feat: Auto retrieval support Element as input (#856)

Wendong-Fan50c93fa7ff25
diff --git a/camel/retrievers/auto_retriever.py b/camel/retrievers/auto_retriever.py
index b84679f7f..4cbc2c4bc 100644
--- a/camel/retrievers/auto_retriever.py
+++ b/camel/retrievers/auto_retriever.py
@@ -28,6 +28,11 @@ from camel.storages import (
)
from camel.types import StorageType
+try:
+ from unstructured.documents.elements import Element
+except ImportError:
+ Element = None
+
DEFAULT_TOP_K_RESULTS = 1
DEFAULT_SIMILARITY_THRESHOLD = 0.75
@@ -97,15 +102,19 @@ class AutoRetriever:
f"Unsupported vector storage type: {self.storage_type}"
)
- def _collection_name_generator(self, content: str) -> str:
+ def _collection_name_generator(self, content: Union[str, Element]) -> str:
r"""Generates a valid collection name from a given file path or URL.
Args:
- contents (str): Local file path, remote URL or string content.
+ content (Union[str, Element]): Local file path, remote URL,
+ string content or Element object.
Returns:
str: A sanitized, valid collection name suitable for use.
"""
+ if isinstance(content, Element):
+ content = content.metadata.file_directory
+
# Check if the content is URL
parsed_url = urlparse(content)
is_url = all([parsed_url.scheme, parsed_url.netloc])
@@ -193,7 +202,7 @@ class AutoRetriever:
def run_vector_retriever(
self,
query: str,
- contents: Union[str, List[str]],
+ contents: Union[str, List[str], Element, List[Element]],
top_k: int = DEFAULT_TOP_K_RESULTS,
similarity_threshold: float = DEFAULT_SIMILARITY_THRESHOLD,
return_detailed_info: bool = False,
@@ -203,8 +212,8 @@ class AutoRetriever:
Args:
query (str): Query string for information retriever.
- contents (Union[str, List[str]]): Local file paths, remote URLs or
- string contents.
+ contents (Union[str, List[str], Element, List[Element]]): Local
+ file paths, remote URLs, string contents or Element objects.
top_k (int, optional): The number of top results to return during
retrieve. Must be a positive integer. Defaults to
`DEFAULT_TOP_K_RESULTS`.
@@ -230,7 +239,9 @@ class AutoRetriever:
if not contents:
raise ValueError("content cannot be empty.")
- contents = [contents] if isinstance(contents, str) else contents
+ contents = (
+ [contents] if isinstance(contents, (str, Element)) else contents
+ )
all_retrieved_info = []
for content in contents:
@@ -246,6 +257,7 @@ class AutoRetriever:
file_is_modified = False # initialize with a default value
if (
vector_storage_instance.status().vector_count != 0
+ and isinstance(content, str)
and os.path.exists(content)
):
# Get original modified date from file
diff --git a/camel/retrievers/vector_retriever.py b/camel/retrievers/vector_retriever.py
index 082c4370e..5ba1d2ef3 100644
--- a/camel/retrievers/vector_retriever.py
+++ b/camel/retrievers/vector_retriever.py
@@ -13,7 +13,7 @@
# =========== Copyright 2023 @ CAMEL-AI.org. All Rights Reserved. ===========
import os
import warnings
-from typing import Any, Dict, List, Optional
+from typing import Any, Dict, List, Optional, Union
from urllib.parse import urlparse
from camel.embeddings import BaseEmbedding, OpenAIEmbedding
@@ -26,6 +26,11 @@ from camel.storages import (
VectorRecord,
)
+try:
+ from unstructured.documents.elements import Element
+except ImportError:
+ Element = None
+
DEFAULT_TOP_K_RESULTS = 1
DEFAULT_SIMILARITY_THRESHOLD = 0.75
@@ -69,7 +74,7 @@ class VectorRetriever(BaseRetriever):
def process(
self,
- content: str,
+ content: Union[str, Element],
chunk_type: str = "chunk_by_title",
**kwargs: Any,
) -> None:
@@ -78,18 +83,22 @@ class VectorRetriever(BaseRetriever):
vector storage.
Args:
- contents (str): Local file path, remote URL or string content.
+ content (Union[str, Element]): Local file path, remote URL,
+ string content or Element object.
chunk_type (str): Type of chunking going to apply. Defaults to
"chunk_by_title".
**kwargs (Any): Additional keyword arguments for content parsing.
"""
- # Check if the content is URL
- parsed_url = urlparse(content)
- is_url = all([parsed_url.scheme, parsed_url.netloc])
- if is_url or os.path.exists(content):
- elements = self.uio.parse_file_or_url(content, **kwargs)
+ if isinstance(content, Element):
+ elements = [content]
else:
- elements = [self.uio.create_element_from_text(text=content)]
+ # Check if the content is URL
+ parsed_url = urlparse(content)
+ is_url = all([parsed_url.scheme, parsed_url.netloc])
+ if is_url or os.path.exists(content):
+ elements = self.uio.parse_file_or_url(content, **kwargs) or []
+ else:
+ elements = [self.uio.create_element_from_text(text=content)]
if elements:
chunks = self.uio.chunk_elements(
chunk_type=chunk_type, elements=elements
@@ -110,7 +119,12 @@ class VectorRetriever(BaseRetriever):
# Prepare the payload for each vector record, includes the content
# path, chunk metadata, and chunk text
for vector, chunk in zip(batch_vectors, batch_chunks):
- content_path_info = {"content path": content}
+ if isinstance(content, str):
+ content_path_info = {"content path": content}
+ elif isinstance(content, Element):
+ content_path_info = {
+ "content path": content.metadata.file_directory
+ }
chunk_metadata = {"metadata": chunk.metadata.to_dict()}
chunk_text = {"text": str(chunk)}
combined_dict = {
diff --git a/test/retrievers/test_auto_retriever.py b/test/retrievers/test_auto_retriever.py
index c41324d0d..21f04237c 100644
--- a/test/retrievers/test_auto_retriever.py
+++ b/test/retrievers/test_auto_retriever.py
@@ -18,6 +18,7 @@ from unittest.mock import patch
import pytest
+from camel.loaders import UnstructuredIO
from camel.retrievers import AutoRetriever
from camel.storages import QdrantStorage
from camel.types import StorageType
@@ -100,3 +101,38 @@ def test_run_vector_retriever(auto_retriever):
)
assert "No suitable information retrieved from" in str(result_unrelated)
+
+
+def test_run_vector_retriever_with_element_input(auto_retriever):
+ uio = UnstructuredIO()
+ test_element = uio.create_element_from_text(
+ text="""Introducing 🦀 CRAB: Cross-environment Agent Benchmark for
+ Multimodal Language Model Agents
+
+ 🦀 CRAB provides an end-to-end and easy-to-use framework to build
+ multimodal agents, operate environments, and create benchmarks to evaluate
+ them, featuring three key components:
+
+ - 🔀 Cross-environment support - agents can operate tasks in 📱 Android
+ and 💻 Ubuntu.
+ - 🕸️ Graph evaluator - provides a fine-grain evaluation metric for agents.
+ - 🤖 Task generation - composes subtasks to automatically generate tasks.
+
+ By connecting all devices to agents, 🦀CRAB unlocks greater capabilities
+ for human-like tasks than ever before.
+
+ Use 🦀 CRAB to benchmark your multimodal agents! """,
+ file_directory="https://x.com/CamelAIOrg/status/1821970132606058943",
+ )
+
+ output = auto_retriever.run_vector_retriever(
+ query="CRAB provides an end-to-end and easy-to-use framework to build"
+ " multimodal agents",
+ contents=test_element,
+ return_detailed_info=True,
+ )
+
+ assert (
+ output["Retrieved Context"][0]['content path']
+ == 'https://x.com/CamelAIOrg/status/1821970132606058943'
+ )