NeMo-Speech.cpp/docs/development/README.md at main · NVIDIA/NeMo-Speech.cpp

NVIDIA · GitHub

Developer guide

Implementation and performance notes for contributors. End users configuring the server want

ASR configuration

,

TTS configuration

, or

Server configuration

instead.

Contents

diagnostics.md

- check_backend_coverage, catching silent CPU fallbacks on a new backend.

asr-batching.md

- exact-shape neural microbatching and indexed streaming-state arenas.

ggml-patches.md

- the project-specific ggml patches and how they are applied at build setup.

cublas-shim.md

- the in-tree drop-in cuBLAS replacement under kernels/ and where the custom GPU kernels live.

Windows build notes