paragekbote/gemma3-torchao-quant-sparse
An optimized gemma-3-4b setup with INT8 weight-only quantization, torch_compile and sparsity for efficient inference.
Public
12
runs