7 projects
bolo-torch
PyTorch engine plugin for the bolo speech server: Indic Parler-TTS, IndicF5 and F5-TTS, Veena, Svara, Indic-Speak, SepFormer, pyannote PixIT and DiCoW diarization-conditioned recognition, run out of process.
bolo-speech
A self-hosted speech server without PyTorch: speech-to-text, text-to-speech, diarization and voice agents (OpenAI audio and Realtime APIs, phone calls, Wyoming), with a model catalog, optional cloud providers and a web console.
uccharan
Phonemiser, text normaliser, transliterator and pronunciation engine for text-to-speech and speech recognition: 39 languages, Apache-2.0.
triton-portcheck
Compile Triton kernels across GPU targets and Triton versions without a GPU, and report where each combination breaks.
dacs-fsspec
fsspec filesystem for DACS: read Azure Blob data through the Distributed Accelerator for Cloud Storage cache from pandas, PyTorch, Hugging Face, Dask and anything else that speaks fsspec
llm-dialects
Streaming parsers for the tool-call and reasoning dialects of open-weight LLMs. Zero dependencies.
engine-migrate
Translate LLM serving configs between inference engines, diff engine versions, and lint dashboards against an engine's metrics - from a database extracted statically from engine source.
