7 lines
324 B
Plaintext
7 lines
324 B
Plaintext
AI Model Dynamic Offloader (AIMDO) is a PyTorch VRAM allocator that implements
|
|
on-demand offloading of model weights when the primary PyTorch VRAM allocator
|
|
comes under pressure.
|
|
|
|
This enables running larger models on GPUs with limited VRAM by intelligently
|
|
managing memory and offloading weights to system RAM when needed.
|