@gsep/adapters-llm-ollama

Ollama adapter for GSEP — use local LLMs with self-evolving prompts

No API keys needed. Run any open-source model locally.

Installation

npm install @gsep/core @gsep/adapters-llm-ollama

Prerequisites

Install and run Ollama:

# Install Ollama, then:
ollama pull llama3
ollama serve

Usage

import { PGA } from '@gsep/core';
import { OllamaAdapter } from '@gsep/adapters-llm-ollama';

const pga = new PGA({
  llm: new OllamaAdapter({
    model: 'llama3',
  }),
});

Configuration

| Option | Type | Default | Description | |--------|------|---------|-------------| | model | string | required | Ollama model name | | baseURL | string | 'http://localhost:11434' | Ollama server URL | | temperature | number | 0.7 | Temperature (0-2) | | timeout | number | 120000 | Request timeout (ms) |

Supported Models

Any model available in Ollama:

• llama3 / llama3.1 / llama3.2
• mistral / mixtral
• deepseek-r1
• phi3 / phi4
• qwen2 / qwen2.5
• gemma2
• And hundreds more

Remote Ollama

Connect to Ollama running on a remote server:

const llm = new OllamaAdapter({
  model: 'llama3',
  baseURL: 'http://your-gpu-server:11434',
});

License

MIT