Python
# InstantHMR distillation training — pip dependencies.
# On Linux, the default PyPI torch wheel already bundles CUDA, so plain
# `torch`/`torchvision` give GPU-enabled builds. If the node needs a specific
# CUDA version, install torch separately from https://pytorch.org first.
# --- core training ---
torch>=2.2
torchvision>=0.17
timm>=1.0
# --- data / image IO ---
numpy>=1.24
pillow>=10
opencv-python>=4.8
# --- LeVJEPA encoder (train_levjepa_*.py) ---
# THIS is what was missing when the perelha job died with
# "ModuleNotFoundError: No module named 'transformers'".
# The weights ship their own modeling code, hence trust_remote_code=True, which
# needs `transformers` itself plus safetensors to read the checkpoint.
transformers>=4.44
safetensors>=0.4
huggingface_hub>=0.24
# --- utilities / logging ---
tqdm>=4.65
matplotlib>=3.7
# --- ONNX export + static quantization (Cell 15) ---
onnx>=1.15
# CPU build is enough for export/quantization. Swap for `onnxruntime-gpu` if
# you also want GPU ONNX inference benchmarking on the node.
onnxruntime>=1.19
# NOTE: the MHR model (checkpoints/mhr_model.pt) is a self-contained TorchScript
# file loaded via torch.jit.load — it does NOT require the `mhr`/`pymomentum`
# packages at runtime, so they are intentionally omitted here.