Upload folder using huggingface_hub
Browse files- README.md +12 -19
- config.json +366 -0
- model.rknn +1 -1
- model_b4_s256.rknn +1 -1
- model_b4_s512.rknn +1 -1
- rknn/model_o1.rknn +1 -1
- rknn/model_o2.rknn +1 -1
- rknn/model_o3.rknn +1 -1
- rknn/model_w8a8.rknn +1 -1
README.md
CHANGED
|
@@ -27,7 +27,7 @@ model_name: ms-marco-MiniLM-L12-v2
|
|
| 27 |
- **Original Model:** [cross-encoder/ms-marco-MiniLM-L12-v2](https://huggingface.co/cross-encoder/ms-marco-MiniLM-L12-v2)
|
| 28 |
- **Target Platform:** rk3588
|
| 29 |
- **rknn-toolkit2 Version:** 2.3.2
|
| 30 |
-
- **rk-transformers Version:** 0.
|
| 31 |
|
| 32 |
### Available Model Files
|
| 33 |
|
|
@@ -46,40 +46,32 @@ model_name: ms-marco-MiniLM-L12-v2
|
|
| 46 |
|
| 47 |
### Installation
|
| 48 |
|
| 49 |
-
Install `rk-transformers` to use this model:
|
| 50 |
|
| 51 |
```bash
|
| 52 |
-
pip install rk-transformers
|
| 53 |
```
|
| 54 |
|
| 55 |
-
####
|
| 56 |
|
| 57 |
```python
|
| 58 |
-
from rktransformers import
|
| 59 |
from transformers import AutoTokenizer
|
| 60 |
|
| 61 |
-
# Load tokenizer and model
|
| 62 |
tokenizer = AutoTokenizer.from_pretrained("rk-transformers/ms-marco-MiniLM-L12-v2")
|
| 63 |
-
model =
|
| 64 |
"rk-transformers/ms-marco-MiniLM-L12-v2",
|
| 65 |
platform="rk3588",
|
| 66 |
core_mask="auto",
|
| 67 |
)
|
| 68 |
|
| 69 |
-
|
| 70 |
-
inputs = tokenizer(
|
| 71 |
-
["Sample text for encoding"],
|
| 72 |
-
padding="max_length",
|
| 73 |
-
max_length=512,
|
| 74 |
-
truncation=True,
|
| 75 |
-
return_tensors="np"
|
| 76 |
-
)
|
| 77 |
-
|
| 78 |
outputs = model(**inputs)
|
| 79 |
-
|
|
|
|
| 80 |
|
| 81 |
# Load specific optimized/quantized model file
|
| 82 |
-
model =
|
| 83 |
"rk-transformers/ms-marco-MiniLM-L12-v2",
|
| 84 |
platform="rk3588",
|
| 85 |
file_name="rknn/model_w8a8.rknn"
|
|
@@ -88,10 +80,11 @@ model = RKRTModelForSequenceClassification.from_pretrained(
|
|
| 88 |
|
| 89 |
## Configuration
|
| 90 |
|
| 91 |
-
The full configuration for all exported RKNN models is available in the [
|
| 92 |
|
| 93 |
</details>
|
| 94 |
|
|
|
|
| 95 |
# Cross-Encoder for MS Marco
|
| 96 |
|
| 97 |
This model was trained on the [MS Marco Passage Ranking](https://github.com/microsoft/MSMARCO-Passage-Ranking) task.
|
|
|
|
| 27 |
- **Original Model:** [cross-encoder/ms-marco-MiniLM-L12-v2](https://huggingface.co/cross-encoder/ms-marco-MiniLM-L12-v2)
|
| 28 |
- **Target Platform:** rk3588
|
| 29 |
- **rknn-toolkit2 Version:** 2.3.2
|
| 30 |
+
- **rk-transformers Version:** 0.3.0
|
| 31 |
|
| 32 |
### Available Model Files
|
| 33 |
|
|
|
|
| 46 |
|
| 47 |
### Installation
|
| 48 |
|
| 49 |
+
Install `rk-transformers` with inference dependencies to use this model:
|
| 50 |
|
| 51 |
```bash
|
| 52 |
+
pip install rk-transformers[inference]
|
| 53 |
```
|
| 54 |
|
| 55 |
+
#### RK-Transformers API
|
| 56 |
|
| 57 |
```python
|
| 58 |
+
from rktransformers import RKModelForSequenceClassification
|
| 59 |
from transformers import AutoTokenizer
|
| 60 |
|
|
|
|
| 61 |
tokenizer = AutoTokenizer.from_pretrained("rk-transformers/ms-marco-MiniLM-L12-v2")
|
| 62 |
+
model = RKModelForSequenceClassification.from_pretrained(
|
| 63 |
"rk-transformers/ms-marco-MiniLM-L12-v2",
|
| 64 |
platform="rk3588",
|
| 65 |
core_mask="auto",
|
| 66 |
)
|
| 67 |
|
| 68 |
+
inputs = tokenizer("Hello, my dog is cute", return_tensors="np")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 69 |
outputs = model(**inputs)
|
| 70 |
+
logits = outputs.logits
|
| 71 |
+
print(logits.shape)
|
| 72 |
|
| 73 |
# Load specific optimized/quantized model file
|
| 74 |
+
model = RKModelForSequenceClassification.from_pretrained(
|
| 75 |
"rk-transformers/ms-marco-MiniLM-L12-v2",
|
| 76 |
platform="rk3588",
|
| 77 |
file_name="rknn/model_w8a8.rknn"
|
|
|
|
| 80 |
|
| 81 |
## Configuration
|
| 82 |
|
| 83 |
+
The full configuration for all exported RKNN models is available in the [config.json](./config.json) file.
|
| 84 |
|
| 85 |
</details>
|
| 86 |
|
| 87 |
+
---
|
| 88 |
# Cross-Encoder for MS Marco
|
| 89 |
|
| 90 |
This model was trained on the [MS Marco Passage Ranking](https://github.com/microsoft/MSMARCO-Passage-Ranking) task.
|
config.json
CHANGED
|
@@ -23,6 +23,372 @@
|
|
| 23 |
"num_hidden_layers": 12,
|
| 24 |
"pad_token_id": 0,
|
| 25 |
"position_embedding_type": "absolute",
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 26 |
"sbert_ce_default_activation_function": "torch.nn.modules.linear.Identity",
|
| 27 |
"torch_dtype": "float32",
|
| 28 |
"transformers_version": "4.55.4",
|
|
|
|
| 23 |
"num_hidden_layers": 12,
|
| 24 |
"pad_token_id": 0,
|
| 25 |
"position_embedding_type": "absolute",
|
| 26 |
+
"rknn": {
|
| 27 |
+
"model.rknn": {
|
| 28 |
+
"batch_size": 1,
|
| 29 |
+
"custom_string": null,
|
| 30 |
+
"dynamic_input": null,
|
| 31 |
+
"float_dtype": "float16",
|
| 32 |
+
"inputs_yuv_fmt": null,
|
| 33 |
+
"max_seq_length": 512,
|
| 34 |
+
"mean_values": null,
|
| 35 |
+
"model_input_names": [
|
| 36 |
+
"input_ids",
|
| 37 |
+
"attention_mask",
|
| 38 |
+
"token_type_ids"
|
| 39 |
+
],
|
| 40 |
+
"opset": 19,
|
| 41 |
+
"optimization": {
|
| 42 |
+
"compress_weight": false,
|
| 43 |
+
"enable_flash_attention": true,
|
| 44 |
+
"model_pruning": false,
|
| 45 |
+
"optimization_level": 0,
|
| 46 |
+
"remove_reshape": false,
|
| 47 |
+
"remove_weight": false,
|
| 48 |
+
"sparse_infer": false
|
| 49 |
+
},
|
| 50 |
+
"quantization": {
|
| 51 |
+
"auto_hybrid_cos_thresh": 0.98,
|
| 52 |
+
"auto_hybrid_euc_thresh": null,
|
| 53 |
+
"dataset_columns": null,
|
| 54 |
+
"dataset_name": null,
|
| 55 |
+
"dataset_size": 128,
|
| 56 |
+
"dataset_split": null,
|
| 57 |
+
"dataset_subset": null,
|
| 58 |
+
"do_quantization": false,
|
| 59 |
+
"quant_img_RGB2BGR": false,
|
| 60 |
+
"quantized_algorithm": "normal",
|
| 61 |
+
"quantized_dtype": "w8a8",
|
| 62 |
+
"quantized_hybrid_level": 0,
|
| 63 |
+
"quantized_method": "channel"
|
| 64 |
+
},
|
| 65 |
+
"rktransformers_version": "0.3.0",
|
| 66 |
+
"single_core_mode": false,
|
| 67 |
+
"std_values": null,
|
| 68 |
+
"target_platform": "rk3588",
|
| 69 |
+
"task": "sequence-classification",
|
| 70 |
+
"task_kwargs": null
|
| 71 |
+
},
|
| 72 |
+
"model_b1_s256.rknn": {
|
| 73 |
+
"batch_size": 1,
|
| 74 |
+
"custom_string": null,
|
| 75 |
+
"dynamic_input": null,
|
| 76 |
+
"float_dtype": "float16",
|
| 77 |
+
"inputs_yuv_fmt": null,
|
| 78 |
+
"max_seq_length": 256,
|
| 79 |
+
"mean_values": null,
|
| 80 |
+
"model_input_names": [
|
| 81 |
+
"input_ids",
|
| 82 |
+
"attention_mask",
|
| 83 |
+
"token_type_ids"
|
| 84 |
+
],
|
| 85 |
+
"opset": 19,
|
| 86 |
+
"optimization": {
|
| 87 |
+
"compress_weight": false,
|
| 88 |
+
"enable_flash_attention": true,
|
| 89 |
+
"model_pruning": false,
|
| 90 |
+
"optimization_level": 0,
|
| 91 |
+
"remove_reshape": false,
|
| 92 |
+
"remove_weight": false,
|
| 93 |
+
"sparse_infer": false
|
| 94 |
+
},
|
| 95 |
+
"quantization": {
|
| 96 |
+
"auto_hybrid_cos_thresh": 0.98,
|
| 97 |
+
"auto_hybrid_euc_thresh": null,
|
| 98 |
+
"dataset_columns": null,
|
| 99 |
+
"dataset_name": null,
|
| 100 |
+
"dataset_size": 128,
|
| 101 |
+
"dataset_split": null,
|
| 102 |
+
"dataset_subset": null,
|
| 103 |
+
"do_quantization": false,
|
| 104 |
+
"quant_img_RGB2BGR": false,
|
| 105 |
+
"quantized_algorithm": "normal",
|
| 106 |
+
"quantized_dtype": "w8a8",
|
| 107 |
+
"quantized_hybrid_level": 0,
|
| 108 |
+
"quantized_method": "channel"
|
| 109 |
+
},
|
| 110 |
+
"rktransformers_version": "0.3.0",
|
| 111 |
+
"single_core_mode": false,
|
| 112 |
+
"std_values": null,
|
| 113 |
+
"target_platform": "rk3588",
|
| 114 |
+
"task": "sequence-classification",
|
| 115 |
+
"task_kwargs": null
|
| 116 |
+
},
|
| 117 |
+
"model_b4_s256.rknn": {
|
| 118 |
+
"batch_size": 4,
|
| 119 |
+
"custom_string": null,
|
| 120 |
+
"dynamic_input": null,
|
| 121 |
+
"float_dtype": "float16",
|
| 122 |
+
"inputs_yuv_fmt": null,
|
| 123 |
+
"max_seq_length": 256,
|
| 124 |
+
"mean_values": null,
|
| 125 |
+
"model_input_names": [
|
| 126 |
+
"input_ids",
|
| 127 |
+
"attention_mask",
|
| 128 |
+
"token_type_ids"
|
| 129 |
+
],
|
| 130 |
+
"opset": 19,
|
| 131 |
+
"optimization": {
|
| 132 |
+
"compress_weight": false,
|
| 133 |
+
"enable_flash_attention": true,
|
| 134 |
+
"model_pruning": false,
|
| 135 |
+
"optimization_level": 0,
|
| 136 |
+
"remove_reshape": false,
|
| 137 |
+
"remove_weight": false,
|
| 138 |
+
"sparse_infer": false
|
| 139 |
+
},
|
| 140 |
+
"quantization": {
|
| 141 |
+
"auto_hybrid_cos_thresh": 0.98,
|
| 142 |
+
"auto_hybrid_euc_thresh": null,
|
| 143 |
+
"dataset_columns": null,
|
| 144 |
+
"dataset_name": null,
|
| 145 |
+
"dataset_size": 128,
|
| 146 |
+
"dataset_split": null,
|
| 147 |
+
"dataset_subset": null,
|
| 148 |
+
"do_quantization": false,
|
| 149 |
+
"quant_img_RGB2BGR": false,
|
| 150 |
+
"quantized_algorithm": "normal",
|
| 151 |
+
"quantized_dtype": "w8a8",
|
| 152 |
+
"quantized_hybrid_level": 0,
|
| 153 |
+
"quantized_method": "channel"
|
| 154 |
+
},
|
| 155 |
+
"rktransformers_version": "0.3.0",
|
| 156 |
+
"single_core_mode": false,
|
| 157 |
+
"std_values": null,
|
| 158 |
+
"target_platform": "rk3588",
|
| 159 |
+
"task": "sequence-classification",
|
| 160 |
+
"task_kwargs": null
|
| 161 |
+
},
|
| 162 |
+
"model_b4_s512.rknn": {
|
| 163 |
+
"batch_size": 4,
|
| 164 |
+
"custom_string": null,
|
| 165 |
+
"dynamic_input": null,
|
| 166 |
+
"float_dtype": "float16",
|
| 167 |
+
"inputs_yuv_fmt": null,
|
| 168 |
+
"max_seq_length": 512,
|
| 169 |
+
"mean_values": null,
|
| 170 |
+
"model_input_names": [
|
| 171 |
+
"input_ids",
|
| 172 |
+
"attention_mask",
|
| 173 |
+
"token_type_ids"
|
| 174 |
+
],
|
| 175 |
+
"opset": 19,
|
| 176 |
+
"optimization": {
|
| 177 |
+
"compress_weight": false,
|
| 178 |
+
"enable_flash_attention": true,
|
| 179 |
+
"model_pruning": false,
|
| 180 |
+
"optimization_level": 0,
|
| 181 |
+
"remove_reshape": false,
|
| 182 |
+
"remove_weight": false,
|
| 183 |
+
"sparse_infer": false
|
| 184 |
+
},
|
| 185 |
+
"quantization": {
|
| 186 |
+
"auto_hybrid_cos_thresh": 0.98,
|
| 187 |
+
"auto_hybrid_euc_thresh": null,
|
| 188 |
+
"dataset_columns": null,
|
| 189 |
+
"dataset_name": null,
|
| 190 |
+
"dataset_size": 128,
|
| 191 |
+
"dataset_split": null,
|
| 192 |
+
"dataset_subset": null,
|
| 193 |
+
"do_quantization": false,
|
| 194 |
+
"quant_img_RGB2BGR": false,
|
| 195 |
+
"quantized_algorithm": "normal",
|
| 196 |
+
"quantized_dtype": "w8a8",
|
| 197 |
+
"quantized_hybrid_level": 0,
|
| 198 |
+
"quantized_method": "channel"
|
| 199 |
+
},
|
| 200 |
+
"rktransformers_version": "0.3.0",
|
| 201 |
+
"single_core_mode": false,
|
| 202 |
+
"std_values": null,
|
| 203 |
+
"target_platform": "rk3588",
|
| 204 |
+
"task": "sequence-classification",
|
| 205 |
+
"task_kwargs": null
|
| 206 |
+
},
|
| 207 |
+
"rknn/model_o1.rknn": {
|
| 208 |
+
"batch_size": 1,
|
| 209 |
+
"custom_string": null,
|
| 210 |
+
"dynamic_input": null,
|
| 211 |
+
"float_dtype": "float16",
|
| 212 |
+
"inputs_yuv_fmt": null,
|
| 213 |
+
"max_seq_length": 512,
|
| 214 |
+
"mean_values": null,
|
| 215 |
+
"model_input_names": [
|
| 216 |
+
"input_ids",
|
| 217 |
+
"attention_mask",
|
| 218 |
+
"token_type_ids"
|
| 219 |
+
],
|
| 220 |
+
"opset": 19,
|
| 221 |
+
"optimization": {
|
| 222 |
+
"compress_weight": false,
|
| 223 |
+
"enable_flash_attention": true,
|
| 224 |
+
"model_pruning": false,
|
| 225 |
+
"optimization_level": 1,
|
| 226 |
+
"remove_reshape": false,
|
| 227 |
+
"remove_weight": false,
|
| 228 |
+
"sparse_infer": false
|
| 229 |
+
},
|
| 230 |
+
"quantization": {
|
| 231 |
+
"auto_hybrid_cos_thresh": 0.98,
|
| 232 |
+
"auto_hybrid_euc_thresh": null,
|
| 233 |
+
"dataset_columns": null,
|
| 234 |
+
"dataset_name": null,
|
| 235 |
+
"dataset_size": 128,
|
| 236 |
+
"dataset_split": null,
|
| 237 |
+
"dataset_subset": null,
|
| 238 |
+
"do_quantization": false,
|
| 239 |
+
"quant_img_RGB2BGR": false,
|
| 240 |
+
"quantized_algorithm": "normal",
|
| 241 |
+
"quantized_dtype": "w8a8",
|
| 242 |
+
"quantized_hybrid_level": 0,
|
| 243 |
+
"quantized_method": "channel"
|
| 244 |
+
},
|
| 245 |
+
"rktransformers_version": "0.3.0",
|
| 246 |
+
"single_core_mode": false,
|
| 247 |
+
"std_values": null,
|
| 248 |
+
"target_platform": "rk3588",
|
| 249 |
+
"task": "sequence-classification",
|
| 250 |
+
"task_kwargs": null
|
| 251 |
+
},
|
| 252 |
+
"rknn/model_o2.rknn": {
|
| 253 |
+
"batch_size": 1,
|
| 254 |
+
"custom_string": null,
|
| 255 |
+
"dynamic_input": null,
|
| 256 |
+
"float_dtype": "float16",
|
| 257 |
+
"inputs_yuv_fmt": null,
|
| 258 |
+
"max_seq_length": 512,
|
| 259 |
+
"mean_values": null,
|
| 260 |
+
"model_input_names": [
|
| 261 |
+
"input_ids",
|
| 262 |
+
"attention_mask",
|
| 263 |
+
"token_type_ids"
|
| 264 |
+
],
|
| 265 |
+
"opset": 19,
|
| 266 |
+
"optimization": {
|
| 267 |
+
"compress_weight": false,
|
| 268 |
+
"enable_flash_attention": true,
|
| 269 |
+
"model_pruning": false,
|
| 270 |
+
"optimization_level": 2,
|
| 271 |
+
"remove_reshape": false,
|
| 272 |
+
"remove_weight": false,
|
| 273 |
+
"sparse_infer": false
|
| 274 |
+
},
|
| 275 |
+
"quantization": {
|
| 276 |
+
"auto_hybrid_cos_thresh": 0.98,
|
| 277 |
+
"auto_hybrid_euc_thresh": null,
|
| 278 |
+
"dataset_columns": null,
|
| 279 |
+
"dataset_name": null,
|
| 280 |
+
"dataset_size": 128,
|
| 281 |
+
"dataset_split": null,
|
| 282 |
+
"dataset_subset": null,
|
| 283 |
+
"do_quantization": false,
|
| 284 |
+
"quant_img_RGB2BGR": false,
|
| 285 |
+
"quantized_algorithm": "normal",
|
| 286 |
+
"quantized_dtype": "w8a8",
|
| 287 |
+
"quantized_hybrid_level": 0,
|
| 288 |
+
"quantized_method": "channel"
|
| 289 |
+
},
|
| 290 |
+
"rktransformers_version": "0.3.0",
|
| 291 |
+
"single_core_mode": false,
|
| 292 |
+
"std_values": null,
|
| 293 |
+
"target_platform": "rk3588",
|
| 294 |
+
"task": "sequence-classification",
|
| 295 |
+
"task_kwargs": null
|
| 296 |
+
},
|
| 297 |
+
"rknn/model_o3.rknn": {
|
| 298 |
+
"batch_size": 1,
|
| 299 |
+
"custom_string": null,
|
| 300 |
+
"dynamic_input": null,
|
| 301 |
+
"float_dtype": "float16",
|
| 302 |
+
"inputs_yuv_fmt": null,
|
| 303 |
+
"max_seq_length": 512,
|
| 304 |
+
"mean_values": null,
|
| 305 |
+
"model_input_names": [
|
| 306 |
+
"input_ids",
|
| 307 |
+
"attention_mask",
|
| 308 |
+
"token_type_ids"
|
| 309 |
+
],
|
| 310 |
+
"opset": 19,
|
| 311 |
+
"optimization": {
|
| 312 |
+
"compress_weight": false,
|
| 313 |
+
"enable_flash_attention": true,
|
| 314 |
+
"model_pruning": false,
|
| 315 |
+
"optimization_level": 3,
|
| 316 |
+
"remove_reshape": false,
|
| 317 |
+
"remove_weight": false,
|
| 318 |
+
"sparse_infer": false
|
| 319 |
+
},
|
| 320 |
+
"quantization": {
|
| 321 |
+
"auto_hybrid_cos_thresh": 0.98,
|
| 322 |
+
"auto_hybrid_euc_thresh": null,
|
| 323 |
+
"dataset_columns": null,
|
| 324 |
+
"dataset_name": null,
|
| 325 |
+
"dataset_size": 128,
|
| 326 |
+
"dataset_split": null,
|
| 327 |
+
"dataset_subset": null,
|
| 328 |
+
"do_quantization": false,
|
| 329 |
+
"quant_img_RGB2BGR": false,
|
| 330 |
+
"quantized_algorithm": "normal",
|
| 331 |
+
"quantized_dtype": "w8a8",
|
| 332 |
+
"quantized_hybrid_level": 0,
|
| 333 |
+
"quantized_method": "channel"
|
| 334 |
+
},
|
| 335 |
+
"rktransformers_version": "0.3.0",
|
| 336 |
+
"single_core_mode": false,
|
| 337 |
+
"std_values": null,
|
| 338 |
+
"target_platform": "rk3588",
|
| 339 |
+
"task": "sequence-classification",
|
| 340 |
+
"task_kwargs": null
|
| 341 |
+
},
|
| 342 |
+
"rknn/model_w8a8.rknn": {
|
| 343 |
+
"batch_size": 1,
|
| 344 |
+
"custom_string": null,
|
| 345 |
+
"dynamic_input": null,
|
| 346 |
+
"float_dtype": "float16",
|
| 347 |
+
"inputs_yuv_fmt": null,
|
| 348 |
+
"max_seq_length": 512,
|
| 349 |
+
"mean_values": null,
|
| 350 |
+
"model_input_names": [
|
| 351 |
+
"input_ids",
|
| 352 |
+
"attention_mask",
|
| 353 |
+
"token_type_ids"
|
| 354 |
+
],
|
| 355 |
+
"opset": 19,
|
| 356 |
+
"optimization": {
|
| 357 |
+
"compress_weight": false,
|
| 358 |
+
"enable_flash_attention": true,
|
| 359 |
+
"model_pruning": false,
|
| 360 |
+
"optimization_level": 0,
|
| 361 |
+
"remove_reshape": false,
|
| 362 |
+
"remove_weight": false,
|
| 363 |
+
"sparse_infer": false
|
| 364 |
+
},
|
| 365 |
+
"quantization": {
|
| 366 |
+
"auto_hybrid_cos_thresh": 0.98,
|
| 367 |
+
"auto_hybrid_euc_thresh": null,
|
| 368 |
+
"dataset_columns": [
|
| 369 |
+
"answer"
|
| 370 |
+
],
|
| 371 |
+
"dataset_name": "sentence-transformers/natural-questions",
|
| 372 |
+
"dataset_size": 1024,
|
| 373 |
+
"dataset_split": [
|
| 374 |
+
"train"
|
| 375 |
+
],
|
| 376 |
+
"dataset_subset": null,
|
| 377 |
+
"do_quantization": true,
|
| 378 |
+
"quant_img_RGB2BGR": false,
|
| 379 |
+
"quantized_algorithm": "normal",
|
| 380 |
+
"quantized_dtype": "w8a8",
|
| 381 |
+
"quantized_hybrid_level": 0,
|
| 382 |
+
"quantized_method": "channel"
|
| 383 |
+
},
|
| 384 |
+
"rktransformers_version": "0.3.0",
|
| 385 |
+
"single_core_mode": false,
|
| 386 |
+
"std_values": null,
|
| 387 |
+
"target_platform": "rk3588",
|
| 388 |
+
"task": "sequence-classification",
|
| 389 |
+
"task_kwargs": null
|
| 390 |
+
}
|
| 391 |
+
},
|
| 392 |
"sbert_ce_default_activation_function": "torch.nn.modules.linear.Identity",
|
| 393 |
"torch_dtype": "float32",
|
| 394 |
"transformers_version": "4.55.4",
|
model.rknn
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 72099070
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:646a50d03c63c0aa2745c6716cf5a25f79fa3b1ee39bd3d266d7fcf074e5e4d8
|
| 3 |
size 72099070
|
model_b4_s256.rknn
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 78763262
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:cd6cd277b1941fbf89bf4cc6e25eaeb19d5781a8e31ccabdd19d4d2433c81ceb
|
| 3 |
size 78763262
|
model_b4_s512.rknn
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 85670846
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:dd83142741f61ae6c8cde672d694dae638509eec65046a1fb9694a5b5944b779
|
| 3 |
size 85670846
|
rknn/model_o1.rknn
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 72099070
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:bfc6b8c69a4dc9e31e698b91e3404f0abe6acf746837e938ca43fce4a31841ac
|
| 3 |
size 72099070
|
rknn/model_o2.rknn
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 72099070
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:b5ca3683f33587cf746c282a72a01778ffa4ec721bc0f996318aadb7957e1deb
|
| 3 |
size 72099070
|
rknn/model_o3.rknn
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 72099070
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:100dbcec0e54ff61cff2b3ebfa52aefa352e7dddb4e50fe8a45deb5390e81282
|
| 3 |
size 72099070
|
rknn/model_w8a8.rknn
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 38286411
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:d1db6b461c724bfecba727a2e268cd1ae637af37aca93adf3def0a4b9e03ee93
|
| 3 |
size 38286411
|