Changelog

This page records LlamaPi features, behavior changes, and fixes.

llamapi-cli and llamapi-server always use the same version number.

0.2.0

Added

Model Deployment and Execution

• Added model discovery, download, loading, execution, and unloading.
• Added interactive chat, single prompts, system prompts, generation-parameter updates, and multiline input to the llamapi-cli.
• Added runtime model IDs, multiple instances, instance resizing, and request queues.

Model Management

• Added local-model, loaded-model, and runtime-state inspection.
• Added removal of local model files.
• Added enable and disable commands for automatic model loading.

Platform and Model Support

• Added automatic model-variant selection based on current hardware and local model availability.
• Added rknn3 and rkllm chat platforms and the rknn2 Embedding platform.
• Added Hugging Face and ModelScope sources with concurrent automatic probing.

API Services

• Added OpenAI-compatible Chat Completions, Embeddings, and Models APIs.
• Added extension APIs for model loading, resizing, unloading, and platform discovery.
• Added regular JSON and SSE streaming responses for Chat Completions.
• Added protocol structures for text, image, audio, and function-tool content.

Configuration and Operations

• Added the llamapi-server systemd service and TOML configuration.
• Added a unified 9265 service-port configuration.

Changed

• None.

Fixed

• None.