Configuration reference for ONNX Runtime Generative AI models, defining model architecture and search parameters.
Field note
This page documents the genai_config.json configuration file used by ONNX Runtime GenAI, covering model architecture settings (type, token IDs, encoder/decoder/vision/speech submodels) and generation search parameters (temperature, top_k, beams, etc.).
Capabilities
No use cases filed yet.
Tags
Ways to use it
No integrations filed yet.
Product features
Define model type, token IDs, vocabulary size, context length, and submodules.
Supports decoder-only, encoder-decoder, and multi-modal model types.
Configure generation search settings like temperature, top_k, beams, and penalties.
Configure ONNX Runtime session options like log_id and provider options.
Supports encoder, decoder, embedding, vision, and speech submodules.