|
47 | 47 | }, |
48 | 48 | "dataloader_config": { |
49 | 49 | "type": "md_quantize_data_loader", |
50 | | - "data_num": 200, |
51 | | - "point_p": 0.3, |
52 | | - "mask_p": 0.2 |
| 50 | + "data_num": 500, |
| 51 | + "point_p": 0.0, |
| 52 | + "mask_p": 0.0 |
53 | 53 | } |
54 | 54 | } |
55 | 55 | ], |
|
66 | 66 | "/Equal_7", "/Unsqueeze_7", "/Add_1", "/Where_7", |
67 | 67 | "/Equal_8", "/Unsqueeze_8", "/Add_2", "/Where_8", |
68 | 68 | "/Equal_9", "/Unsqueeze_9", "/Add_3", "/Where_9", |
69 | | - "/Equal_10", "/Unsqueeze_10", "/Add_4", "/Where_10"] |
| 69 | + "/Equal_10", "/Unsqueeze_10", "/Add_4", "/Where_10", |
| 70 | + "/transformer/layers.0/norm1/LayerNormalization"] |
70 | 71 | }, |
71 | 72 | "sq": { |
72 | 73 | "type": "OnnxStaticQuantization", |
73 | 74 | "data_config": "quantize_data_config", |
74 | 75 | "activation_type": "uint8", |
75 | 76 | "precision": "uint8", |
76 | 77 | "calibration_providers": [ "CUDAExecutionProvider" ], |
77 | | - "calibrate_method": "MinMax", |
| 78 | + "calibrate_method": "Percentile", |
78 | 79 | "quant_preprocess": true, |
79 | 80 | "op_types_to_exclude": ["Cast"], |
80 | 81 | "save_as_external_data": true, |
|
86 | 87 | "/Equal_7", "/Unsqueeze_7", "/Add_1", "/Where_7", |
87 | 88 | "/Equal_8", "/Unsqueeze_8", "/Add_2", "/Where_8", |
88 | 89 | "/Equal_9", "/Unsqueeze_9", "/Add_3", "/Where_9", |
89 | | - "/Equal_10", "/Unsqueeze_10", "/Add_4", "/Where_10"] |
| 90 | + "/Equal_10", "/Unsqueeze_10", "/Add_4", "/Where_10", |
| 91 | + "/transformer/layers.0/norm1/LayerNormalization"] |
90 | 92 | }, |
91 | 93 | "cb": { |
92 | 94 | "type": "EPContextBinaryGenerator", |
|
0 commit comments