Dockerfile for Demo with 1.7B Base Model

[iamantonreznik]

This is a Dockerfile for quickly building and running the demo interface.

Tested on Ubuntu 22.04.5 LTS with an RTX 3090.

Use it for Demo with Voice Cloning and TTS.

Usage:

docker build -t demo -f demo/dockerfiles/Dockerfile-1.7B-Base .
docker run --rm --gpus all -p 8000:8000 demo

Uses pytorch:2.8.0-cuda12.8-cudnn9-runtime and wheel for flash attention.
Image size aprox. 4.6 GB.

[bryanstevensacosta]

Nice addition! Adding a Dockerfile to quickly spin up the demo with the 1.7B base model is a great improvement for developer experience and testing.

A few suggestions that could help polish it before merging:

• Usage clarity: It might be helpful to slightly expand the README with a minimal docker run example, including which port is exposed and which endpoint can be used to test TTS inference.
• Environment variables: Nice touch adding HF_HOME. It could be useful to briefly explain why it’s needed and whether there are other recommended environment variables (ex. for model caching or credentials).
• Version compatibility: Consider documenting the expected PyTorch/CUDA versions (either in comments or README) to make reproduction more consistent across different hosts and GPUs.
• Health check (Optional): A lightweight sanity check or inference test during container startup could help ensure the model loads correctly.