ollama-proxy

ollama-proxy is a service that adheres to the Ollama external interface protocol. It enables connections to various large model services, such as glm-4-plus and deepseek-coder-v2, through the Ollama interface protocol. The primary goal of this project is to facilitate integration with zed.dev, allowing it to utilize additional large model services beyond the official configuration.

Installation

To install ollama-proxy, a command-line tool, you can use the following command:

pip install ollama-proxy

Command Line Usage

The ollama-proxy command-line interface (CLI) allows you to run the proxy server with specific configurations. Below are the details on how to use the CLI, including available options and their descriptions.

Command Structure

To run the ollama-proxy, use the following command structure:

ollama-proxy <model_name> --config <config_path> --host <host> --port <port>

Arguments

• model_name:

  • Type: String
  • Description: A required argument that specifies the name of the model you want to run.

• config:

  • Type: String
  • Default: keys.toml
  • Description: The path to the TOML configuration file.

Options

• --host:

  • Type: String
  • Default: 127.0.0.1
  • Description: The host address for the server. For zed.dev, only localhost is allowed.

• --port:

  • Type: Integer
  • Default: 11434
  • Description: For zed.dev, the port number must be 11434.

Example Usage

To start the ollama-proxy with a specific model and configuration, you can run:

ollama-proxy glm-4-plus --config path/to/config.toml --host localhost --port 11434 --reload

In this example:

• glm-4-plus is the name of the model you want to run.
• The configuration file is located at path/to/config.toml.
• The server will listen on localhost at port 11434.
• Hot reloading is enabled.

Notes

• Ensure that the specified configuration file exists and is correctly formatted in TOML.
• The server will print logs to the console, indicating its status and any errors that may occur.

Configuration TOML File

To configure ollama-proxy, create a configuration file in TOML format. This file allows you to specify various settings for the proxy.

Example Configuration (config.toml)

[glm-4-plus]
provider = "zhipu"
url = "https://open.bigmodel.cn/api/paas/v4/chat/completions"
api_key = "your_api_key_here"

[deepseek-coder-v2]
provider = "deepseek"
url = "https://api.deepseek.com/chat/completions"
api_key = "your_api_key_here"

• The section name corresponds to the model name, such as glm-4-plus or deepseek-coder-v2.
• The provider specifies the name of the large model service provider.
• The url is the endpoint of the large model service provider.
• The api_key is the API key for the large model service provider.

When you run ollama-proxy, it will load the configuration file. For example, running ollama-proxy glm-4-plus will use glm-4-plus as the running model name. If glm-4-plus is not specified in the configuration file, an error will be raised.

TODO

• Support more large model providers
• Add xinference as a local model provider
• Support cursor