moondream2-torchao-W4A8 / quant_meta.json
Azaz666's picture
Upload torchao-W4A8 quantized model
c5b8872 verified
{
"model_id": "vikhyatk/moondream2",
"family": "moondream",
"method": "torchao_w4a8",
"bits_weight": 4,
"bits_activation": 8,
"group_size": 128,
"skip_vision": true,
"load_time_s": 8.6,
"quant_time_s": 0.2,
"quant_method": "Int8DynamicActivationInt4Weight (W4A8)"
}