perf(tts): Magpie TTS - CUDA/CPU fast path - #51

Merged

pskrunner14 merged 13 commits into

mainfrom

perf/magpie-cuda

Sep 30, 2026

Commits on Sep 29, 2026

perf(tts): faster MagpieTTS CUDA decode path
pskrunner14
committed
c261a1f
perf(tts): prefetch text-chunk transitions, Gumbel sampler, lock-free prefetch path
pskrunner14
committed
b3dac22
fix(tts): portable MagpieTTS fast paths
pskrunner14
committed
bc7d287
perf(tts): fused MagpieTTS kernels on mid-size GPUs, bench tts metrics
pskrunner14
committed
e017c6d
fix(tts): address review comments
pskrunner14
committed
9514e6d
docs(tts): refresh MagpieTTS benchmark results | update coderabbit.yaml instructions
pskrunner14
committed
5719339
fix(cuda): make skinny-Q8 dispatch independent of request history
pskrunner14
committed
809c4d1
docs: ASR benchmarks and NeMo reference results
pskrunner14
committed
cc95207
docs: add AI usage guidelines to CONTRIBUTING.md
pskrunner14
committed
08ef11e

Commits on Sep 30, 2026