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: gemini embedding support. (#2325)

LastWhispere7fe81b8ef9d
diff --git a/camel/embeddings/__init__.py b/camel/embeddings/__init__.py
index 420c0db20..79c464c8f 100644
--- a/camel/embeddings/__init__.py
+++ b/camel/embeddings/__init__.py
@@ -13,6 +13,7 @@
# ========= Copyright 2023-2024 @ CAMEL-AI.org. All Rights Reserved. =========
from .azure_embedding import AzureEmbedding
from .base import BaseEmbedding
+from .gemini_embedding import GeminiEmbedding
from .jina_embedding import JinaEmbedding
from .mistral_embedding import MistralEmbedding
from .openai_compatible_embedding import OpenAICompatibleEmbedding
@@ -31,4 +32,5 @@ __all__ = [
"OpenAICompatibleEmbedding",
"JinaEmbedding",
"TogetherEmbedding",
+ "GeminiEmbedding",
]
diff --git a/camel/embeddings/gemini_embedding.py b/camel/embeddings/gemini_embedding.py
new file mode 100644
index 000000000..7dbeb56a3
--- /dev/null
+++ b/camel/embeddings/gemini_embedding.py
@@ -0,0 +1,115 @@
+# ========= Copyright 2023-2024 @ CAMEL-AI.org. All Rights Reserved. =========
+# 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.
+# ========= Copyright 2023-2024 @ CAMEL-AI.org. All Rights Reserved. =========
+from __future__ import annotations
+
+import os
+from typing import Any, Optional
+
+from camel.embeddings.base import BaseEmbedding
+from camel.types import EmbeddingModelType, GeminiEmbeddingTaskType
+from camel.utils import api_keys_required
+
+
+class GeminiEmbedding(BaseEmbedding[str]):
+ r"""Provides text embedding functionalities using Google's Gemini models.
+
+ Args:
+ model_type (EmbeddingModelType, optional): The model type to be
+ used for text embeddings.
+ (default: :obj:`GEMINI_EMBEDDING_EXP`)
+ api_key (str, optional): The API key for authenticating with the
+ Gemini service. (default: :obj:`None`)
+ dimensions (int, optional): The text embedding output dimensions.
+ (default: :obj:`None`)
+ task_type (GeminiEmbeddingTaskType, optional): The task type for which
+ to optimize the embeddings. (default: :obj:`None`)
+
+ Raises:
+ RuntimeError: If an unsupported model type is specified.
+ """
+
+ @api_keys_required(
+ [
+ ("api_key", 'GEMINI_API_KEY'),
+ ]
+ )
+ def __init__(
+ self,
+ model_type: EmbeddingModelType = (
+ EmbeddingModelType.GEMINI_EMBEDDING_EXP
+ ),
+ api_key: Optional[str] = None,
+ dimensions: Optional[int] = None,
+ task_type: Optional[GeminiEmbeddingTaskType] = None,
+ ) -> None:
+ from google import genai
+
+ if not model_type.is_gemini:
+ raise ValueError("Invalid Gemini embedding model type.")
+
+ self.model_type = model_type
+ if dimensions is None:
+ self.output_dim = model_type.output_dim
+ else:
+ assert isinstance(dimensions, int)
+ self.output_dim = dimensions
+
+ self._api_key = api_key or os.environ.get("GEMINI_API_KEY")
+ self._task_type = task_type
+
+ # Initialize Gemini client
+ self._client = genai.Client(api_key=self._api_key)
+
+ def embed_list(
+ self,
+ objs: list[str],
+ **kwargs: Any,
+ ) -> list[list[float]]:
+ r"""Generates embeddings for the given texts.
+
+ Args:
+ objs (list[str]): The texts for which to generate the embeddings.
+ **kwargs (Any): Extra kwargs passed to the embedding API.
+
+ Returns:
+ list[list[float]]: A list that represents the generated embedding
+ as a list of floating-point numbers.
+ """
+ from google.genai import types
+
+ # Create embedding config if task_type is specified
+ embed_config = None
+ if self._task_type:
+ embed_config = types.EmbedContentConfig(
+ task_type=self._task_type.value
+ )
+
+ # Process each text separately since Gemini API
+ # expects single content item
+ responses = self._client.models.embed_content(
+ model=self.model_type.value,
+ contents=objs, # type: ignore[arg-type]
+ config=embed_config,
+ **kwargs,
+ )
+
+ return [response.values for response in responses.embeddings] # type: ignore[misc,union-attr]
+
+ def get_output_dim(self) -> int:
+ r"""Returns the output dimension of the embeddings.
+
+ Returns:
+ int: The dimensionality of the embedding for the current model.
+ """
+ return self.output_dim
diff --git a/camel/embeddings/openai_embedding.py b/camel/embeddings/openai_embedding.py
index a708d78a9..2bf77a749 100644
--- a/camel/embeddings/openai_embedding.py
+++ b/camel/embeddings/openai_embedding.py
@@ -14,7 +14,7 @@
from __future__ import annotations
import os
-from typing import Any
+from typing import Any, Optional, Union
from openai import OpenAI
@@ -27,15 +27,15 @@ class OpenAIEmbedding(BaseEmbedding[str]):
r"""Provides text embedding functionalities using OpenAI's models.
Args:
- model_type (EmbeddingModelType, optional): The model type to be
- used for text embeddings.
+ model_type (EmbeddingModelType): The model type to be used for text
+ embeddings.
(default: :obj:`TEXT_EMBEDDING_3_SMALL`)
- url (Optional[str], optional): The url to the OpenAI service.
+ url (Optional[str]): The url to the OpenAI service.
(default: :obj:`None`)
- api_key (str, optional): The API key for authenticating with the
- OpenAI service. (default: :obj:`None`)
- dimensions (int, optional): The text embedding output dimensions.
- (default: :obj:`NOT_GIVEN`)
+ api_key (Optional[str]): The API key for authenticating with
+ the OpenAI service. (default: :obj:`None`)
+ dimensions (Union[int, NotGiven]): The text embedding output
+ dimensions. (default: :obj:`NOT_GIVEN`)
Raises:
RuntimeError: If an unsupported model type is specified.
@@ -51,9 +51,9 @@ class OpenAIEmbedding(BaseEmbedding[str]):
model_type: EmbeddingModelType = (
EmbeddingModelType.TEXT_EMBEDDING_3_SMALL
),
- url: str | None = None,
- api_key: str | None = None,
- dimensions: int | NotGiven = NOT_GIVEN,
+ url: Optional[str] = None,
+ api_key: Optional[str] = None,
+ dimensions: Union[int, NotGiven] = NOT_GIVEN,
) -> None:
if not model_type.is_openai:
raise ValueError("Invalid OpenAI embedding model type.")
diff --git a/camel/embeddings/together_embedding.py b/camel/embeddings/together_embedding.py
index 4880a66fb..e95f3529d 100644
--- a/camel/embeddings/together_embedding.py
+++ b/camel/embeddings/together_embedding.py
@@ -27,7 +27,7 @@ class TogetherEmbedding(BaseEmbedding[str]):
r"""Provides text embedding functionalities using Together AI's models.
Args:
- model_name (str, optional): The model name to be used for text
+ model_type (str, optional): The model name to be used for text
embeddings.
(default: :obj:`togethercomputer/m2-bert-80M-8k-retrieval`)
api_key (str, optional): The API key for authenticating with the
@@ -44,17 +44,17 @@ class TogetherEmbedding(BaseEmbedding[str]):
@api_keys_required([("api_key", 'TOGETHER_API_KEY')])
def __init__(
self,
- model_name: str = "togethercomputer/m2-bert-80M-8k-retrieval",
+ model_type: str = "togethercomputer/m2-bert-80M-8k-retrieval",
api_key: Optional[str] = None,
dimensions: Optional[int] = None,
) -> None:
- if not isinstance(model_name, str) or not model_name.strip():
+ if not isinstance(model_type, str) or not model_type.strip():
raise ValueError("Model name must be a non-empty string")
if dimensions is not None and dimensions <= 0:
raise ValueError("Dimensions must be a positive integer")
- self.model_name = model_name
+ self.model_type = model_type
self._api_key = api_key or os.environ.get("TOGETHER_API_KEY")
self.output_dim = dimensions
@@ -91,7 +91,7 @@ class TogetherEmbedding(BaseEmbedding[str]):
try:
response = self.client.embeddings.create(
input=objs,
- model=self.model_name,
+ model=self.model_type,
**kwargs,
)
@@ -100,7 +100,7 @@ class TogetherEmbedding(BaseEmbedding[str]):
self.output_dim = len(response.data[0].embedding)
logger.debug(
f"Set output dimension to {self.output_dim} for model "
- f"{self.model_name}"
+ f"{self.model_type}"
)
return [data.embedding for data in response.data]
@@ -130,7 +130,7 @@ class TogetherEmbedding(BaseEmbedding[str]):
if self.output_dim is None:
raise ValueError(
"Failed to determine embedding dimension for model: "
- f"{self.model_name}"
+ f"{self.model_type}"
)
return self.output_dim
diff --git a/camel/types/__init__.py b/camel/types/__init__.py
index 8a0729e2a..27d1c4902 100644
--- a/camel/types/__init__.py
+++ b/camel/types/__init__.py
@@ -14,6 +14,7 @@
from .enums import (
AudioModelType,
EmbeddingModelType,
+ GeminiEmbeddingTaskType,
HuggingFaceRepoType,
ModelPlatformType,
ModelType,
@@ -75,6 +76,7 @@ __all__ = [
'UnifiedModelType',
'ParsedChatCompletion',
'HuggingFaceRepoType',
+ 'GeminiEmbeddingTaskType',
'NOT_GIVEN',
'NotGiven',
]
diff --git a/camel/types/enums.py b/camel/types/enums.py
index 9c3679858..72e856209 100644
--- a/camel/types/enums.py
+++ b/camel/types/enums.py
@@ -1202,6 +1202,8 @@ class EmbeddingModelType(Enum):
MISTRAL_EMBED = "mistral-embed"
+ GEMINI_EMBEDDING_EXP = "gemini-embedding-exp-03-07"
+
@property
def is_openai(self) -> bool:
r"""Returns whether this type of models is an OpenAI-released model."""
@@ -1230,6 +1232,13 @@ class EmbeddingModelType(Enum):
EmbeddingModelType.MISTRAL_EMBED,
}
+ @property
+ def is_gemini(self) -> bool:
+ r"""Returns whether this type of models is an Gemini-released model."""
+ return self in {
+ EmbeddingModelType.GEMINI_EMBEDDING_EXP,
+ }
+
@property
def output_dim(self) -> int:
if self in {
@@ -1253,10 +1262,29 @@ class EmbeddingModelType(Enum):
return 3072
elif self is EmbeddingModelType.MISTRAL_EMBED:
return 1024
+ elif self is EmbeddingModelType.GEMINI_EMBEDDING_EXP:
+ return 3072
else:
raise ValueError(f"Unknown model type {self}.")
+class GeminiEmbeddingTaskType(str, Enum):
+ r"""Task types for Gemini embedding models.
+
+ For more information, please refer to:
+ https://ai.google.dev/gemini-api/docs/embeddings#task-types
+ """
+
+ SEMANTIC_SIMILARITY = "SEMANTIC_SIMILARITY"
+ CLASSIFICATION = "CLASSIFICATION"
+ CLUSTERING = "CLUSTERING"
+ RETRIEVAL_DOCUMENT = "RETRIEVAL_DOCUMENT"
+ RETRIEVAL_QUERY = "RETRIEVAL_QUERY"
+ QUESTION_ANSWERING = "QUESTION_ANSWERING"
+ FACT_VERIFICATION = "FACT_VERIFICATION"
+ CODE_RETRIEVAL_QUERY = "CODE_RETRIEVAL_QUERY"
+
+
class TaskType(Enum):
AI_SOCIETY = "ai_society"
CODE = "code"
diff --git a/docs/key_modules/embeddings.md b/docs/key_modules/embeddings.md
index 73d74d9ad..c70278b71 100644
--- a/docs/key_modules/embeddings.md
+++ b/docs/key_modules/embeddings.md
@@ -109,7 +109,7 @@ embeddings = azure_openai_embedding.embed_list(["Hello, world!", "Another exampl
from camel.embeddings import TogetherEmbedding
# Initialize the Together AI embedding with a specific model
-together_embedding = TogetherEmbedding(model_name="togethercomputer/m2-bert-80M-8k-retrieval")
+together_embedding = TogetherEmbedding(model_type="togethercomputer/m2-bert-80M-8k-retrieval")
# Generate embeddings for a list of texts
embeddings = together_embedding.embed_list(["Hello, world!", "Another example"])
diff --git a/examples/embeddings/gemini_embedding_example.py b/examples/embeddings/gemini_embedding_example.py
new file mode 100644
index 000000000..f80f84cd7
--- /dev/null
+++ b/examples/embeddings/gemini_embedding_example.py
@@ -0,0 +1,30 @@
+# ========= Copyright 2023-2024 @ CAMEL-AI.org. All Rights Reserved. =========
+# 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.
+# ========= Copyright 2023-2024 @ CAMEL-AI.org. All Rights Reserved. =========
+
+from camel.embeddings import GeminiEmbedding
+from camel.types import EmbeddingModelType
+
+gemini_embed = GeminiEmbedding(
+ model_type=EmbeddingModelType.GEMINI_EMBEDDING_EXP
+)
+
+# Embed a single text
+text_embeddings = gemini_embed.embed_list(["What is the capital of France?"])
+
+print("Default model embedding dimension:", len(text_embeddings[0]))
+'''
+===============================================================================
+Default model embedding dimension: 3072
+===============================================================================
+'''
diff --git a/examples/embeddings/together_embedding_example.py b/examples/embeddings/together_embedding_example.py
index 69f20db35..9cca5eb13 100644
--- a/examples/embeddings/together_embedding_example.py
+++ b/examples/embeddings/together_embedding_example.py
@@ -28,7 +28,7 @@ Default model embedding dimension: 768
# Example with a different model
together_embed_alt = TogetherEmbedding(
- model_name="togethercomputer/m2-bert-80M-2k-retrieval",
+ model_type="togethercomputer/m2-bert-80M-2k-retrieval",
)
# Embed multiple texts
diff --git a/pyproject.toml b/pyproject.toml
index 81554d09f..bf7842eac 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -78,6 +78,7 @@ rag = [
"pandasai>=2.3.0,<3",
"crawl4ai>=0.3.745",
"pyobvector>=0.1.18",
+ "google-genai>=1.13.0",
]
web_tools = [
"duckduckgo-search>=6.3.5,<7",
@@ -359,6 +360,7 @@ all = [
"scrapegraph-py>=1.12.0,<2",
"daytona-sdk==0.14.0",
"ibm-watsonx-ai>=1.3.11",
+ "google-genai>=1.13.0",
]
[project.urls]
@@ -548,6 +550,7 @@ module = [
"daytona_sdk",
"scrapegraph_py.*",
"ibm_watsonx_ai.*",
+ "google_genai",
]
ignore_missing_imports = true
diff --git a/test/embeddings/test_gemini_embedding.py b/test/embeddings/test_gemini_embedding.py
new file mode 100644
index 000000000..44f647f5f
--- /dev/null
+++ b/test/embeddings/test_gemini_embedding.py
@@ -0,0 +1,101 @@
+# ========= Copyright 2023-2024 @ CAMEL-AI.org. All Rights Reserved. =========
+# 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.
+# ========= Copyright 2023-2024 @ CAMEL-AI.org. All Rights Reserved. =========
+import os
+from unittest.mock import MagicMock, patch
+
+from camel.embeddings import GeminiEmbedding
+from camel.types import GeminiEmbeddingTaskType
+
+
[email protected](os.environ, {"GEMINI_API_KEY": "fake_api_key"})
+@patch('google.genai.Client.models', autospec=True)
+def test_embed_list(mock_models):
+ # Set up the mock models and its return values
+ mock_embed_content = MagicMock()
+ mock_models.embed_content = mock_embed_content
+
+ # Create two different responses for the two texts
+ mock_response = MagicMock()
+ mock_embedding1 = MagicMock()
+ mock_embedding1.values = [0.1, 0.2, 0.3]
+ mock_embedding2 = MagicMock()
+ mock_embedding2.values = [0.4, 0.5, 0.6]
+ mock_response.embeddings = [mock_embedding1, mock_embedding2]
+
+ # Set up the mock to return different responses for each call
+ mock_embed_content.side_effect = [mock_response]
+
+ # Create the embedding instance and test embed_list
+ embedding = GeminiEmbedding()
+ result = embedding.embed_list(["text1", "text2"])
+
+ # Verify the calls to the mock
+ assert mock_embed_content.call_count == 1
+
+ # Check the first call
+ args1, kwargs1 = mock_embed_content.call_args_list[0]
+ assert kwargs1['model'] == "gemini-embedding-exp-03-07"
+ assert kwargs1['contents'] == ["text1", "text2"]
+ assert kwargs1['config'] is None
+
+ # Verify the result
+ assert result == [[0.1, 0.2, 0.3], [0.4, 0.5, 0.6]]
+
+
[email protected](os.environ, {"GEMINI_API_KEY": "fake_api_key"})
+@patch('google.genai.Client.models', autospec=True)
+def test_embed_list_with_task_type(mock_models):
+ # Set up the mock models and its return values
+ mock_embed_content = MagicMock()
+ mock_models.embed_content = mock_embed_content
+
+ # Create response for the text
+ mock_response = MagicMock()
+ mock_embedding = MagicMock()
+ mock_embedding.values = [0.1, 0.2, 0.3]
+ mock_response.embeddings = [mock_embedding]
+
+ # Set up the mock to return the response
+ mock_embed_content.return_value = mock_response
+
+ # Create the embedding instance with task type and test embed_list
+ embedding = GeminiEmbedding(
+ task_type=GeminiEmbeddingTaskType.SEMANTIC_SIMILARITY
+ )
+ result = embedding.embed_list(["text1"])
+
+ # Verify the call to the mock includes task_type in config
+ mock_embed_content.assert_called_once()
+ args, kwargs = mock_embed_content.call_args
+ assert kwargs['model'] == "gemini-embedding-exp-03-07"
+ assert kwargs['contents'] == ["text1"]
+ assert kwargs['config'] is not None
+
+ # Verify the result
+ assert result == [[0.1, 0.2, 0.3]]
+
+
[email protected](os.environ, {"GEMINI_API_KEY": "fake_api_key"})
+def test_get_output_dim():
+ # Instantiate the GeminiEmbedding with specified dimensions
+ embedding = GeminiEmbedding(dimensions=3072)
+
+ # Validate that get_output_dim returns the correct value
+ assert embedding.get_output_dim() == 3072
+
+ # Instantiate the GeminiEmbedding without specified dimensions
+ embedding = GeminiEmbedding()
+
+ # Validate that get_output_dim returns the correct value
+ assert embedding.get_output_dim() == 3072
diff --git a/uv.lock b/uv.lock
index 897b1d8f9..79a153ffa 100644
--- a/uv.lock
+++ b/uv.lock
@@ -589,6 +589,7 @@ all = [
{ name = "google-auth-httplib2" },
{ name = "google-auth-oauthlib" },
{ name = "google-cloud-storage" },
+ { name = "google-genai" },
{ name = "googlemaps" },
{ name = "gradio" },
{ name = "html2text" },
@@ -821,6 +822,7 @@ owl = [
rag = [
{ name = "cohere" },
{ name = "crawl4ai" },
+ { name = "google-genai" },
{ name = "nebula3-python" },
{ name = "neo4j" },
{ name = "numpy" },
@@ -968,6 +970,8 @@ requires-dist = [
{ name = "google-auth-oauthlib", marker = "extra == 'web-tools'", specifier = "==1.2.1" },
{ name = "google-cloud-storage", marker = "extra == 'all'", specifier = ">=2.18.0,<3" },
{ name = "google-cloud-storage", marker = "extra == 'storage'", specifier = ">=2.18.0,<3" },
+ { name = "google-genai", marker = "extra == 'all'", specifier = ">=1.13.0" },
+ { name = "google-genai", marker = "extra == 'rag'", specifier = ">=1.13.0" },
{ name = "googlemaps", marker = "extra == 'all'", specifier = ">=4.10.0,<5" },
{ name = "googlemaps", marker = "extra == 'web-tools'", specifier = ">=4.10.0,<5" },
{ name = "gradio", marker = "extra == 'all'", specifier = ">=3,<4" },
@@ -2472,6 +2476,24 @@ wheels = [
{ url = "https://files.pythonhosted.org/packages/fd/3c/2a19a60a473de48717b4efb19398c3f914795b64a96cf3fbe82588044f78/google_crc32c-1.7.1-pp311-pypy311_pp73-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:6efb97eb4369d52593ad6f75e7e10d053cf00c48983f7a973105bc70b0ac4d82", size = 28048 },
]
+[[package]]
+name = "google-genai"
+version = "1.13.0"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+ { name = "anyio" },
+ { name = "google-auth" },
+ { name = "httpx" },
+ { name = "pydantic" },
+ { name = "requests" },
+ { name = "typing-extensions" },
+ { name = "websockets" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/5e/1f/0861dad47febbb391123a529f3655582e2a7ce14aadb6ddc527dd7dea470/google_genai-1.13.0.tar.gz", hash = "sha256:de7b4960d361c7ea85b715ce0e07fe34903d94affa82d566d9eb04e8f92512b9", size = 165727 }
+wheels = [
+ { url = "https://files.pythonhosted.org/packages/c5/2c/1833fa15311fa4f0790e9c8ea3488382fe428c3eca62b329d6aa355871ae/google_genai-1.13.0-py3-none-any.whl", hash = "sha256:e88417c95f333827ed051282803db627192fd8e5fdf5b49b6f26644a67775d1a", size = 164405 },
+]
+
[[package]]
name = "google-resumable-media"
version = "2.7.2"