provider-lm-studio (LM Studio Example with Promptfoo)

You can run this example with:

npx promptfoo@latest init --example provider-lm-studio
cd provider-lm-studio

This example demonstrates how to use Promptfoo with LM Studio for prompt evaluation. It showcases configuration for interacting with the LM Studio API using a locally hosted language model.

Prerequisites

1. LM Studio: Install LM Studio from lmstudio.ai.
2. Model: Download the bartowski/gemma-2-9b-it-GGUF model in LM Studio.

Setup

1. Start LM Studio Server:

   • Open LM Studio and load the bartowski/gemma-2-9b-it-GGUF model.
   • Start a local server to host the model (usually at http://localhost:1234).

2. Configure Promptfoo:

   Create a promptfooconfig.yaml file with the following content:

   yamlCopy

   providers:
     - id: 'http://localhost:1234/v1/chat/completions'
       config:
         method: 'POST'
         headers:
           'Content-Type': 'application/json'
         body:
           messages: '{{ prompt }}'
           model: 'bartowski/gemma-2-9b-it-GGUF'
           temperature: 0.7
           max_tokens: -1
         transformResponse: 'json.choices[0].message.content'
   

   Note that you can view the specific configuration for each model within LM Studio's examples in the server tab.

Usage

1. Run Evaluation:

   bashCopy

   npx promptfoo eval
   

2. View Results:

   shCopy

   npx promptfoo view
   

For more information, see the Promptfoo documentation on how to set up a custom HTTP provider.