DeepSeek_LLM_Use.md

DeepSeek Usage

1. Create the relevant file.

Create a YAML file, for example, user_deepseek_llm.yaml Paste the following content into your user_deepseek_llm.yaml file.

name: 'user_deepseek_llm'
description: 'default default_deepseek_llm llm with spi'
model_name: 'deepseek-chat'
max_tokens: 1000
temperature: 0.5
api_key: '${DEEPSEEK_API_KEY}'
api_base: 'https://api.deepseek.com/v1'
organization: '${DEEPSEEK_ORGANIZATION}'
proxy: '${DEEPSEEK_PROXY}'
streaming: True
metadata:
  type: 'LLM'
  module: 'agentuniverse.llm.default.deep_seek_openai_style_llm'
  class: 'DefaultDeepSeekLLM'

Note:

The model parameters such as api_key, api_base, organization, and proxy can be configured in three ways:

1. Direct String Value: Enter the API key string directly in the configuration file.

   api_key: 'sk-xxx'

2. Environment Variable Placeholder: Use the ${VARIABLE_NAME} syntax to load from environment variables. When agentUniverse is launched, it will automatically read the corresponding value from the environment variables.

   api_key: '${DEEPSEEK_API_KEY}'

3. Custom Function Loading: Use the @FUNC annotation to dynamically load the API key via a custom function at runtime.

   api_key: '@FUNC(load_api_key(model_name="deepseek"))'

   The function needs to be defined in the YamlFuncExtension class within the yaml_func_extension.py file. You can refer to the example in the sample project's YamlFuncExtension. When agentUniverse loads this configuration:

   • It parses the @FUNC annotation.
   • Executes the load_api_key function with the corresponding parameters.
   • Replaces the annotation content with the function's return value.

2. Environment Setup

In the example YAML, model keys and other parameters are configured using environment variable placeholders. The following section will introduce methods for setting environment variables.

Must be configured: DEEPSEEK_API_KEY
Optional: DEEPSEEK_API_BASE, DEEPSEEK_PROXY, DEEPSEEK_ORGANIZATION

2.1 Configure through Python code

import os
os.environ['DEEPSEEK_API_KEY'] = 'sk-***'
os.environ['DEEPSEEK_API_BASE'] = 'https://xxxxxx'

2.2 Configure through the configuration file

In the custom_key.toml file under the config directory of the project, add the configuration:

DEEPSEEK_API_KEY="sk-******"
DEEPSEEK_API_BASE="https://xxxxxx"
DEEPSEEK_ORGANIZATION = ''
DEEPSEEK_PROXY = ''

3. Obtaining the DeepSeek API KEY

Please refer to the official documentation of DEEPSEEK:https://platform.deepseek.com/api_keys

4. Tips

In the agentuniverse, we have established an LLM named default_deepseek_llm. Users can directly utilize it after configuring their DEEPSEEK_API_KEY.