Skip to content

Conversation

@manascb1344
Copy link
Owner

Working GPU based API, run using docker

docker build -t zonos-api-local .
docker run -d \
  --name zonos-api-local \
  --gpus all \
  -p 8000:8000 \
  -e CUDA_VISIBLE_DEVICES=0 \
  zonos-api

or directly in dev mode as:

uvicorn app.main:app --host 0.0.0.0 --port 8000

- Implement main.py with comprehensive TTS API routes
- Add routes for model discovery, language support, and speech synthesis
- Include error handling and streaming response for audio generation
- Update requirements.txt to remove version constraints
- Modify Dockerfile to install Zonos repository directly
- Add new sampling parameters to TTSService: top_p, top_k, linear, confidence, and quadratic
- Update generate method to support expanded sampling configuration
- Create a comprehensive sampling_params dictionary for more granular audio generation control
- Update Dockerfile to install Zonos package directly
- Modify main.py route for model conditioners and add debugging logs
- Enhance models.py with default values and optional parameters
- Refactor tts.py service with improved logging and model initialization
- Add more comprehensive error handling and logging across components
- Update TTSRequest model to include sampling configuration parameters
- Modify main.py to use dynamic sampling parameters from API request
- Add default values for top_p, top_k, linear, confidence, and quadratic parameters
- Make seed parameter non-optional with a default value
- Consolidate pip installations in Dockerfile for better layer caching
- Add --no-cache-dir flag to reduce image size
- Modify COPY command for app directory
- Increase TTS generation max tokens from 30 to 60 seconds
…e dependency installation with uv package manager
@manascb1344 manascb1344 merged commit a999e53 into main Feb 25, 2025
3 of 4 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants