/seamless-m4t-install

seamless m4t installation Guide

MIT LicenseMIT

Seamless M4T for Linux

Seamless M4T Linux installation Guide

This installation guide has been tested on Debian 12 Bookworm on a Lenovo Legion 4 Core i7, 16GB RAM, SSD 1TB, Nvidia GTX 1055 3GB Laptop.

It contains the fundamental steps to be able to launch the Meta Seamless M4T project, I encountered several problems along the way and that's why I developed this guide so that you can jump over the obstacles.

Things I found and didn't work for me:

The Traductor model uses 2 models seamlessM4T_large (10GB), and seamlessm4T_medium (6GB)

The seamlessM4T_large model doesn't work on my PC, it runs out of RAM LOL, so I modified this line to change and use the medium model, if you modify that line it should look like this:

translator = Translator(
    model_name_or_card="seamlessM4T_medium",
    vocoder_name_or_card="vocoder_36langs",
    device=device,
    dtype=torch.float16 if "cuda" in device.type else torch.float32,
)

So, if the process ends and is not able to start the APP then you will have to change it before running the Gradio App, locate the line of code that I left you above and change it in your project (seamless_communication/demo/app.py)

Recommended requirements

  • Nvidia Card with Drivers installed

  • Debian 12|11, Ubuntu 22.04

  • Miniconda 3

  • Computer >= 15GB RAM, Multi Core

Start Process:

install debian libs

sudo apt update -y && sudo apt install git ffmpeg -y

Prepare Conda Environment with Python 3.10

conda create -n mt4 python=3.10
conda activate mt4

Download sources

git clone https://github.com/facebookresearch/seamless_communication
cd seamless_communication

install requirements

pip install .
# optional
# conda install -y -c conda-forge libsndfile
conda install -c conda-forge libsndfile==1.0.31

launch App M4T Gui

cd demo
pip install -r requirements.txt
python app.py share=True

Additional Resources

Nvidia Cuda Linux

https://developer.download.nvidia.com/compute/redist/ffmpeg/1511-patch/FFMPEG-with-NVIDIA-Acceleration-on-Ubuntu_UG_v01.pdf

https://linuxconfig.org/how-to-install-nvidia-driver-on-debian-12-bookworm-linux