mohitsha's picture
First model version
5acb56c
{
"one_external_file": true,
"opset": null,
"optimization": {},
"optimum_version": "1.14.0.dev0",
"quantization": {
"activations_dtype": "QUInt8",
"activations_symmetric": false,
"format": "QDQ",
"is_static": true,
"mode": "QLinearOps",
"nodes_to_exclude": [
"/proj_out/MatMul",
"/model/decoder/0/self_attn/q_proj/MatMul_smooth_mul",
"/model/decoder/0/self_attn/k_proj/MatMul_smooth_mul",
"/model/decoder/0/self_attn/v_proj/MatMul_smooth_mul",
"/model/decoder/0/self_attn/out_proj/MatMul_smooth_mul",
"/model/decoder/0/encoder_attn/q_proj/MatMul_smooth_mul",
"/model/decoder/0/encoder_attn/out_proj/MatMul_smooth_mul",
"/model/decoder/0/fc1/MatMul_smooth_mul",
"/model/decoder/0/fc2/MatMul_smooth_mul",
"/model/decoder/1/self_attn/q_proj/MatMul_smooth_mul",
"/model/decoder/1/self_attn/k_proj/MatMul_smooth_mul",
"/model/decoder/1/self_attn/v_proj/MatMul_smooth_mul",
"/model/decoder/1/self_attn/out_proj/MatMul_smooth_mul",
"/model/decoder/1/encoder_attn/MatMul_smooth_mul",
"/model/decoder/1/encoder_attn/MatMul_5_smooth_mul",
"/model/decoder/1/fc1/MatMul_smooth_mul",
"/model/decoder/1/fc2/MatMul_smooth_mul",
"/model/decoder/2/self_attn/q_proj/MatMul_smooth_mul",
"/model/decoder/2/self_attn/k_proj/MatMul_smooth_mul",
"/model/decoder/2/self_attn/v_proj/MatMul_smooth_mul",
"/model/decoder/2/self_attn/out_proj/MatMul_smooth_mul",
"/model/decoder/2/encoder_attn/MatMul_smooth_mul",
"/model/decoder/2/encoder_attn/MatMul_5_smooth_mul",
"/model/decoder/2/fc1/MatMul_smooth_mul",
"/model/decoder/2/fc2/MatMul_smooth_mul",
"/model/decoder/3/self_attn/q_proj/MatMul_smooth_mul",
"/model/decoder/3/self_attn/k_proj/MatMul_smooth_mul",
"/model/decoder/3/self_attn/v_proj/MatMul_smooth_mul",
"/model/decoder/3/self_attn/out_proj/MatMul_smooth_mul",
"/model/decoder/3/encoder_attn/MatMul_smooth_mul",
"/model/decoder/3/encoder_attn/MatMul_5_smooth_mul",
"/model/decoder/3/fc1/MatMul_smooth_mul",
"/model/decoder/3/fc2/MatMul_smooth_mul",
"/proj_out/MatMul_smooth_mul"
],
"nodes_to_quantize": [],
"operators_to_quantize": [
"MatMul"
],
"per_channel": false,
"qdq_add_pair_to_weight": false,
"qdq_dedicated_pair": false,
"qdq_op_type_per_channel_support_to_axis": {
"MatMul": 1
},
"reduce_range": false,
"smooth_quant": false,
"smooth_quant_alpha": 0.5,
"smooth_quant_folding": true,
"smooth_quant_op_types": [
"MatMul"
],
"weights_dtype": "QInt8",
"weights_symmetric": true
},
"transformers_version": "4.33.1",
"use_external_data_format": false
}