6d5fb65a72
Add environment variables for fine-grained control over TTS rate limiting and Better Auth behavior. Move documentation to external Docusaurus site with automated deployment workflows. - TTS rate limiting can now be enabled/disabled via TTS_ENABLE_RATE_LIMIT - Customizable daily limits for anonymous/authenticated users and IP backstops - Better Auth rate limiting can be disabled via DISABLE_AUTH_RATE_LIMIT - Rename library import env vars to IMPORT_LIBRARY_DIRS/DIR - Add docs-site with Docusaurus and GitHub Actions workflows - Update README to reference external documentation
49 lines
1.3 KiB
Markdown
49 lines
1.3 KiB
Markdown
---
|
|
title: Kokoro-FastAPI
|
|
---
|
|
|
|
You can run the Kokoro TTS API server directly with Docker.
|
|
|
|
:::warning
|
|
For Kokoro issues and support, use the upstream repository: [remsky/Kokoro-FastAPI](https://github.com/remsky/Kokoro-FastAPI).
|
|
:::
|
|
|
|
## CPU image
|
|
|
|
```bash
|
|
docker run -d \
|
|
--name kokoro-tts \
|
|
--restart unless-stopped \
|
|
-p 8880:8880 \
|
|
-e ONNX_NUM_THREADS=8 \
|
|
-e ONNX_INTER_OP_THREADS=4 \
|
|
-e ONNX_EXECUTION_MODE=parallel \
|
|
-e ONNX_OPTIMIZATION_LEVEL=all \
|
|
-e ONNX_MEMORY_PATTERN=true \
|
|
-e ONNX_ARENA_EXTEND_STRATEGY=kNextPowerOfTwo \
|
|
-e API_LOG_LEVEL=DEBUG \
|
|
ghcr.io/remsky/kokoro-fastapi-cpu:v0.2.4
|
|
```
|
|
|
|
## GPU image
|
|
|
|
```bash
|
|
docker run -d \
|
|
--name kokoro-tts \
|
|
--gpus all \
|
|
--user 1001:1001 \
|
|
--restart unless-stopped \
|
|
-p 8880:8880 \
|
|
-e USE_GPU=true \
|
|
-e PYTHONUNBUFFERED=1 \
|
|
-e API_LOG_LEVEL=DEBUG \
|
|
ghcr.io/remsky/kokoro-fastapi-gpu:v0.2.4
|
|
```
|
|
|
|
## OpenReader integration notes
|
|
|
|
- In OpenReader settings, choose provider `Custom OpenAI-Like` and model `Kokoro`.
|
|
- Set OpenReader `API_BASE` to your Kokoro endpoint (for Docker Compose, commonly `http://kokoro-tts:8880/v1`).
|
|
- `API_BASE` is needed here because Kokoro is used via the custom provider path, not a built-in provider endpoint.
|
|
- GPU mode requires NVIDIA Docker support and is best on NVIDIA hardware.
|
|
- CPU mode works best on Apple Silicon or modern x86 CPUs.
|