diff --git a/examples/recipes/intfloat_multilingual-e5-large/cpu/cpu/feature-extraction_fp16_config.json b/examples/recipes/intfloat_multilingual-e5-large/cpu/cpu/feature-extraction_fp16_config.json new file mode 100644 index 000000000..5b339a13d --- /dev/null +++ b/examples/recipes/intfloat_multilingual-e5-large/cpu/cpu/feature-extraction_fp16_config.json @@ -0,0 +1,76 @@ +{ + "export": { + "opset_version": 17, + "batch_size": 1, + "export_params": true, + "do_constant_folding": true, + "verbose": false, + "dynamo": false, + "enable_hierarchy_tags": true, + "clean_onnx": false, + "hierarchy_tag_format": "full", + "input_tensors": [ + { + "name": "input_ids", + "dtype": "int32", + "shape": [ + 1, + 512 + ], + "value_range": [ + 0, + 250002 + ] + }, + { + "name": "attention_mask", + "dtype": "int32", + "shape": [ + 1, + 512 + ], + "value_range": [ + 0, + 2 + ] + } + ], + "output_tensors": [ + { + "name": "last_hidden_state" + } + ] + }, + "optim": { + "clamp_constant_values": true + }, + "quant": { + "mode": "fp16", + "samples": 10, + "calibration_method": "minmax", + "weight_type": "uint8", + "activation_type": "uint8", + "per_channel": false, + "symmetric": false, + "weight_symmetric": null, + "activation_symmetric": null, + "save_calibration": false, + "distribution": "uniform", + "seed": null, + "calibration_load_path": null, + "calibration_save_path": null, + "op_types_to_quantize": null, + "nodes_to_exclude": null, + "task": "feature-extraction", + "model_id": "intfloat/multilingual-e5-large", + "model_type": "xlm-roberta", + "fp16_keep_io_types": true, + "fp16_op_block_list": null + }, + "compile": null, + "loader": { + "task": "feature-extraction", + "model_class": "AutoModel", + "model_type": "xlm-roberta" + } +} diff --git a/examples/recipes/intfloat_multilingual-e5-large/cpu/cpu/feature-extraction_fp32_config.json b/examples/recipes/intfloat_multilingual-e5-large/cpu/cpu/feature-extraction_fp32_config.json new file mode 100644 index 000000000..71e4c3177 --- /dev/null +++ b/examples/recipes/intfloat_multilingual-e5-large/cpu/cpu/feature-extraction_fp32_config.json @@ -0,0 +1,54 @@ +{ + "export": { + "opset_version": 17, + "batch_size": 1, + "export_params": true, + "do_constant_folding": true, + "verbose": false, + "dynamo": false, + "enable_hierarchy_tags": true, + "clean_onnx": false, + "hierarchy_tag_format": "full", + "input_tensors": [ + { + "name": "input_ids", + "dtype": "int32", + "shape": [ + 1, + 512 + ], + "value_range": [ + 0, + 250002 + ] + }, + { + "name": "attention_mask", + "dtype": "int32", + "shape": [ + 1, + 512 + ], + "value_range": [ + 0, + 2 + ] + } + ], + "output_tensors": [ + { + "name": "last_hidden_state" + } + ] + }, + "optim": { + "clamp_constant_values": true + }, + "quant": null, + "compile": null, + "loader": { + "task": "feature-extraction", + "model_class": "AutoModel", + "model_type": "xlm-roberta" + } +}