Drop both caches and reclaim GPU memory. Useful between phases that load many models in one process.
Source code in src/interlens/loading/model_cache.py
| def free() -> None:
"""Drop both caches and reclaim GPU memory. Useful between phases that load many models in one process."""
_WEIGHTS.clear()
_TOKENIZERS.clear()
gc.collect()
if torch.cuda.is_available():
torch.cuda.empty_cache()
|