跳到主要内容
Open In ColabOpen on GitHub

OpenGradientToolkit

本笔记展示了如何使用 OpenGradient 工具包构建工具。此工具包使用户能够基于 OpenGradient 网络上的模型和工作流创建自定义工具。

设置​

请确保您拥有 OpenGradient API 密钥，以便访问 OpenGradient 网络。如果您已有 API 密钥，只需设置环境变量

!export OPENGRADIENT_PRIVATE_KEY="your-api-key"

如果您需要设置新的 API 密钥，请下载 opengradient SDK 并按照说明初始化新配置。

!pip install opengradient
!opengradient config init

安装​

此工具包位于 langchain-opengradient 包中

%pip install -qU langchain-opengradient

实例化​

现在我们可以使用之前获得的 API 密钥实例化我们的工具包。

from langchain_opengradient import OpenGradientToolkit

toolkit = OpenGradientToolkit(
    # Not required if you have already set the environment variable OPENGRADIENT_PRIVATE_KEY
    private_key="your-api-key"
)

构建您自己的工具​

OpenGradientToolkit 提供了两种创建自定义工具的主要方法

1. 创建一个运行机器学习模型的工具​

您可以创建利用部署在 OpenGradient 模型中心上的机器学习模型的工具。用户创建的模型可以通过 OpenGradient SDK 上传、推断并共享到模型中心。

import opengradient as og
from pydantic import BaseModel, Field


# Example 1: Simple tool with no input schema
def price_data_provider():
    """Function that provides input data to the model."""
    return {
        "open_high_low_close": [
            [2535.79, 2535.79, 2505.37, 2515.36],
            [2515.37, 2516.37, 2497.27, 2506.94],
            [2506.94, 2515, 2506.35, 2508.77],
            [2508.77, 2519, 2507.55, 2518.79],
            [2518.79, 2522.1, 2513.79, 2517.92],
            [2517.92, 2521.4, 2514.65, 2518.13],
            [2518.13, 2525.4, 2517.2, 2522.6],
            [2522.59, 2528.81, 2519.49, 2526.12],
            [2526.12, 2530, 2524.11, 2529.99],
            [2529.99, 2530.66, 2525.29, 2526],
        ]
    }


def format_volatility(inference_result):
    """Function that formats the model output."""
    return format(float(inference_result.model_output["Y"].item()), ".3%")


# Create the tool
volatility_tool = toolkit.create_run_model_tool(
    model_cid="QmRhcpDXfYCKsimTmJYrAVM4Bbvck59Zb2onj3MHv9Kw5N",
    tool_name="eth_volatility",
    model_input_provider=price_data_provider,
    model_output_formatter=format_volatility,
    tool_description="Generates volatility measurement for ETH/USDT trading pair",
    inference_mode=og.InferenceMode.VANILLA,
)


# Example 2: Tool with input schema from the agent
class TokenInputSchema(BaseModel):
    token: str = Field(description="Token name (ethereum or bitcoin)")


def token_data_provider(**inputs):
    """Dynamic function that changes behavior based on agent input."""
    token = inputs.get("token")
    if token == "bitcoin":
        return {"price_series": [100001.1, 100013.2, 100149.2, 99998.1]}
    else:  # ethereum
        return {"price_series": [2010.1, 2012.3, 2020.1, 2019.2]}


# Create the tool with schema
token_tool = toolkit.create_run_model_tool(
    model_cid="QmZdSfHWGJyzBiB2K98egzu3MypPcv4R1ASypUxwZ1MFUG",
    tool_name="token_volatility",
    model_input_provider=token_data_provider,
    model_output_formatter=lambda x: format(float(x.model_output["std"].item()), ".3%"),
    tool_input_schema=TokenInputSchema,
    tool_description="Measures return volatility for a specified token",
)

# Add tools to the toolkit
toolkit.add_tool(volatility_tool)
toolkit.add_tool(token_tool)

2. 创建一个读取工作流结果的工具​

读取工作流是定期运行存储在智能合约中模型的调度推断，这些模型使用实时预言机数据。更多信息可在此处找到。

您可以创建从工作流智能合约中读取结果的工具

# Create a tool to read from a workflow
forecast_tool = toolkit.create_read_workflow_tool(
    workflow_contract_address="0x58826c6dc9A608238d9d57a65bDd50EcaE27FE99",
    tool_name="ETH_Price_Forecast",
    tool_description="Reads latest forecast for ETH price from deployed workflow",
    output_formatter=lambda x: f"Price change forecast: {format(float(x.numbers['regression_output'].item()), '.2%')}",
)

# Add the tool to the toolkit
toolkit.add_tool(forecast_tool)

工具​

使用内置的 get_tools() 方法查看 OpenGradient 工具包中可用的工具列表。

tools = toolkit.get_tools()

# View tools
for tool in tools:
    print(tool)

在代理中使用​

以下是如何将 OpenGradient 工具与 LangChain 代理结合使用

from langchain_openai import ChatOpenAI
from langgraph.prebuilt import create_react_agent

# Initialize LLM
llm = ChatOpenAI(model="gpt-4o")

# Create tools from the toolkit
tools = toolkit.get_tools()

# Create agent
agent_executor = create_react_agent(llm, tools)

# Example query for the agent
example_query = "What's the current volatility of ETH?"

# Execute the agent
events = agent_executor.stream(
    {"messages": [("user", example_query)]},
    stream_mode="values",
)
for event in events:
    event["messages"][-1].pretty_print()

以下是所有内容整合后的示例输出

================================ Human Message =================================

What's the current volatility of ETH?
================================== Ai Message ==================================
Tool Calls:
  eth_volatility (chatcmpl-tool-d66ab9ee8f2c40e5a2634d90c7aeb17d)
 Call ID: chatcmpl-tool-d66ab9ee8f2c40e5a2634d90c7aeb17d
  Args:
================================= Tool Message =================================
Name: eth_volatility

0.038%
================================== Ai Message ==================================

The current volatility of the ETH/USDT trading pair is 0.038%.

API 参考​

欲了解更多详情，请参见 Github 页面。