a work in progress
- uv
uv venv <VENV_PATH> --python 3.12source <VENV_PATH>/bin/activateuv pip install torch==2.5.1 torchvision==0.20.1 --index-url https://download.pytorch.org/whl/cu121uv pip install matplotlib==3.10.1 datasets==3.3.2 transformers==4.49.0 scikit-learn==1.6.1 python-dotenv==1.0.1 accelerate==1.4.0
cp .env.sample .env- modify HF_ env variables as needed
python test_env.py
- consumed ~ 3 GB of VRAM
TrainOutput(global_step=1350, training_loss=0.15539069290514346, metrics={'train_runtime': 137.439, 'train_samples_per_second': 98.225, 'train_steps_per_second': 9.823, 'total_flos': 1.046216869705728e+18, 'train_loss': 0.15539069290514346, 'epoch': 3.0}){'test_loss': 0.06985098868608475, 'test_accuracy': 0.978, 'test_runtime': 3.4191, 'test_samples_per_second': 292.471, 'test_steps_per_second': 73.118}-
consumed ~ ? GB of VRAM
-
in progress
-
consumed ~ 7 GB of VRAM
-
please ignore aimv2 results till the correct base classes are used!
TrainOutput(global_step=1350, training_loss=1.853396900318287, metrics={'train_runtime': 494.6235, 'train_samples_per_second': 27.293, 'train_steps_per_second': 2.729, 'total_flos': 2.928337017403392e+18, 'train_loss': 1.853396900318287, 'epoch': 3.0})