Blazingly fast LLM inference.

State-of-the-art audio codec with 90x compression factor. Supports 44.1kHz, 24kHz, and 16kHz mono/stereo audio.

SimVQ: Addressing Representation Collapse in Vector Quantized Models with One Linear Layer

Run GGUF models easily with a KoboldAI UI. One File. Zero Install.

[ICLR 2025] SOTA discrete acoustic codec models with 40/75 tokens per second for audio language modeling

LLM inference in C/C++

Interface for OuteTTS models.