File size: 1,536 Bytes
af7a81c | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 | Fixed parameters:
* **model_name_or_path**: `Bhumika/roberta-base-finetuned-sst2`
* **dataset**:
* **path**: `glue`
* **name**: `sst2`
* **calibration_split**: `None`
* **eval_split**: `validation`
* **data_keys**: `['sentence']`
* **label_keys**: `['label']`
* **quantization_approach**: `dynamic`
* **node_exclusion**: `[]`
* **per_channel**: `False`
* **calibration**: `None`
* **framework**: `onnxruntime`
* **framework_args**:
* **opset**: `15`
* **optimization_level**: `1`
* **aware_training**: `False`
Benchmarked parameters:
* **operators_to_quantize**: `['Add', 'MatMul']`, `['Add']`
## Evaluation
Below, time metrics for
* Batch size: 8
* Input length: 128
| operators_to_quantize | | latency_mean (original, ms) | latency_mean (optimized, ms) | | throughput (original, /s) | throughput (optimized, /s) | | accuracy (original) | accuracy (optimized) |
| :-------------------: | :-: | :-------------------------: | :--------------------------: | :-: | :-----------------------: | :------------------------: | :-: | :-----------------: | :------------------: |
| `['Add']` | \| | 454.70 | 361.81 | \| | 2.50 | 3.00 | \| | 1.0 | 1.0 |
| `['Add', 'MatMul']` | \| | 474.54 | 135.14 | \| | 2.50 | 7.50 | \| | 1.0 | 1.0 |
|