paligemma-3b-mix-224 / quantize_config.json
Xenova's picture
Xenova HF staff
Upload folder using huggingface_hub
69b9d6c verified
raw
history blame
3.22 kB
{
"fp16": {},
"q8": {
"per_model_config": {
"embed_tokens": {
"op_types": [
"Gather"
],
"weight_type": "QInt8"
},
"vision_encoder": {
"op_types": [
"Add",
"Cast",
"Concat",
"Constant",
"Conv",
"Div",
"Gather",
"MatMul",
"Mul",
"Pow",
"ReduceMean",
"Reshape",
"Shape",
"Slice",
"Softmax",
"Sqrt",
"Sub",
"Tanh",
"Transpose",
"Unsqueeze"
],
"weight_type": "QUInt8"
}
},
"per_channel": false,
"reduce_range": false
},
"int8": {
"per_model_config": {
"embed_tokens": {
"op_types": [
"Gather"
],
"weight_type": "QInt8"
},
"vision_encoder": {
"op_types": [
"Add",
"Cast",
"Concat",
"Constant",
"Conv",
"Div",
"Gather",
"MatMul",
"Mul",
"Pow",
"ReduceMean",
"Reshape",
"Shape",
"Slice",
"Softmax",
"Sqrt",
"Sub",
"Tanh",
"Transpose",
"Unsqueeze"
],
"weight_type": "QInt8"
}
},
"per_channel": false,
"reduce_range": false
},
"uint8": {
"per_model_config": {
"embed_tokens": {
"op_types": [
"Gather"
],
"weight_type": "QUInt8"
},
"vision_encoder": {
"op_types": [
"Add",
"Cast",
"Concat",
"Constant",
"Conv",
"Div",
"Gather",
"MatMul",
"Mul",
"Pow",
"ReduceMean",
"Reshape",
"Shape",
"Slice",
"Softmax",
"Sqrt",
"Sub",
"Tanh",
"Transpose",
"Unsqueeze"
],
"weight_type": "QUInt8"
}
},
"per_channel": false,
"reduce_range": false
},
"q4": {
"block_size": 32,
"is_symmetric": true,
"accuracy_level": null
},
"bnb4": {
"block_size": 64,
"quant_type": 1
}
}