NEW VERSION TEST

[nagadit]

The repository is alive, don't worry. Soon I plan to upload a new version of the scripts, as well as scripts for converting models to DFLive and its subsequent launch. Who can help me with script tests on different video cards?

I created a discussion where you will write the results of the work with a full description of the system and python packages

Required:

• RTX 3x series
• RTX 2x series
• GTX 1x series

#Systems:

• Ubuntu 16,18,20
• Debian 10,11,12
• CentOS
• Arch Linux
• Redhat

Let's make DFLab and DFLive easier to use in unix systems :)

[l0ophole]

I'll help anyway I can. My GPU probably is too old for the new version, though, right? [rpf@arch3 ~]$ python3 cuda_check.py Found 1 device(s). Device: 0 Name: NVIDIA GeForce GT 635 Compute Capability: 3.5 Multiprocessors: 2 CUDA Cores: 384 Concurrent threads: 4096 GPU clock: 967 MHz Memory clock: 1001 MHz Total Memory: 980 MiB Free Memory: 593 MiB

[skydam]

Would love to help. Running Ubuntu 22 with RTX 3090. Having trouble with current scripts as it refuses to train on the GPU (only CPU).

[hueylewis85]

I have been using linux Suse, with two 3090 and cudnn above 8 to be faster, the installation I made was:

conda create -n deepfacelab -c main -c "nvidia/label/cuda-11.4.0" -c conda-forge python=3.7 cudnn=8.4.1 cudatoolkit=11.4

modify the file requirements-cuda.txt

tqdm numpy==1.20 numexpr h5py==2.10.0 opencv-python==4.1.0.25 ffmpeg-python==0.1.17 scikit-image==0.14.2 scipy==1.4.1 colorama tensorflow-gpu==2.11 pyqt5 tf2onnx

works with dual 3090 gpu's

[marshalleq]

Well I take it this wasn't fully completed given it's still open - I have a rather extravagant 4090, happy to help test anything. Sort of hoping this comment will revive this thread as it's so far quite difficult to get running on Linux.

[LazyBai]

it's too difficult.....waht settings should i use to run on the 4090?,,,....

[mdhvishv]

I have A6000 gpu and ubuntu 22.04 but facing issues

[mdhvishv]

its blacksreen problem