scott f20699aed3 Add fp16 autocast to synthesis for faster GPU throughput ...

The 6700 XT has significantly higher fp16 throughput than fp32.
autocast("cuda") uses fp16 for matmuls and convolutions (HiFiGAN,
S3 tokenizer, flow matching) while keeping fp32 for precision-sensitive
ops like softmax and layer norm.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

2026-04-05 13:34:21 -04:00

.gitea/workflows

Switch to inline cache to avoid registry blob size limits

2026-04-05 12:14:35 -04:00

.gitignore

Initial implementation: Chatterbox TTS with ROCm and Wyoming

2026-04-05 09:51:09 -04:00

config.py

Initial implementation: Chatterbox TTS with ROCm and Wyoming

2026-04-05 09:51:09 -04:00

config.yaml

Initial implementation: Chatterbox TTS with ROCm and Wyoming

2026-04-05 09:51:09 -04:00

docker-compose.yml

Enable HSA_OVERRIDE_GFX_VERSION=10.3.0 for RX 6700 XT

2026-04-05 13:07:46 -04:00

Dockerfile.rocm

Pin base image to rocm/dev-ubuntu-22.04:7.2, restore torch 2.11.0

2026-04-05 12:35:58 -04:00

engine.py

Add fp16 autocast to synthesis for faster GPU throughput

2026-04-05 13:34:21 -04:00

main.py

Initial implementation: Chatterbox TTS with ROCm and Wyoming

2026-04-05 09:51:09 -04:00

perth_stub.py

Replace resemble-perth with a no-op stub

2026-04-05 11:16:08 -04:00

requirements-rocm-init.txt

Pin base image to rocm/dev-ubuntu-22.04:7.2, restore torch 2.11.0

2026-04-05 12:35:58 -04:00

requirements-rocm.txt

Replace resemble-perth with a no-op stub

2026-04-05 11:16:08 -04:00

wyoming_handler.py

Stream audio in 4096-sample sub-chunks for immediate HA playback

2026-04-05 13:25:54 -04:00

wyoming_voices.py

Fix TtsVoice missing required version argument

2026-04-05 11:52:14 -04:00

Description

No description provided

213 KiB

Languages

Python 100%