许可协议: apache-2.0
数据集:

• librispeech_asr
  评估指标:
• wer
  流水线标签: 自动语音识别
  标签:
• 自动语音识别
• ONNX
• 英特尔®神经压缩器
• neural-compressor
  库名称: transformers

INT4 Whisper large-v2 ONNX模型

Whisper是一个预训练的自动语音识别（ASR）和语音翻译模型。经过68万小时标注数据的训练，Whisper模型展现出强大的泛化能力，无需微调即可适应多种数据集和领域。此仓库包含ONNX格式的Whisper large v2模型的INT4仅权重量化版本，由英特尔®神经压缩器和英特尔®Transformers扩展驱动。

该INT4 ONNX模型由英特尔®神经压缩器的仅权重量化方法生成。

导出为ONNX模型

FP32模型通过openai/whisper-large-v2导出：

optimum-cli export onnx --model openai/whisper-large-v2 whisper-large-v2-with-past/ --task automatic-speech-recognition-with-past --opset 13

安装ONNX Runtime

安装onnxruntime>=1.16.0以支持MatMulFpQ4运算符。

运行量化

从主分支构建英特尔®神经压缩器并运行INT4仅权重量化。

权重量化配置如下：

关键代码如下。完整脚本请参考whisper示例。

from neural_compressor import quantization, PostTrainingQuantConfig
from neural_compressor.utils.constant import FP32

model_list = ['encoder_model.onnx', 'decoder_model.onnx', 'decoder_with_past_model.onnx']
for model in model_list:
    config = PostTrainingQuantConfig(
        approach="weight_only",
        calibration_sampling_size=[8],
        op_type_dict={".*": {"weight": {"bits": 4, 
                                        "algorithm": ["RTN"], 
                                        "scheme": ["sym"], 
                                        "group_size": 32}}},)
    q_model = quantization.fit(
        os.path.join("/path/to/whisper-large-v2-with-past", model), # FP32模型路径
        config,
        calib_dataloader=dataloader)
    q_model.save(os.path.join("/path/to/whisper-large-v2-onnx-int4", model)) # INT4模型路径

评估

运算符统计
INT4 ONNX模型的运算符统计如下：

词错率评估
在librispeech_asr数据集上评估模型：

import os
from evaluate import load
from datasets import load_dataset
from transformers import WhisperForConditionalGeneration, WhisperProcessor, AutoConfig
model_name = 'openai/whisper-large-v2'
model_path = 'whisper-large-v2-onnx-int4'
processor = WhisperProcessor.from_pretrained(model_name)
model = WhisperForConditionalGeneration.from_pretrained(model_name)
config = AutoConfig.from_pretrained(model_name)
wer = load("wer")
librispeech_test_clean = load_dataset("librispeech_asr", "clean", split="test")

from optimum.onnxruntime import ORTModelForSpeechSeq2Seq
from transformers import PretrainedConfig
model_config = PretrainedConfig.from_pretrained(model_name)
predictions = []
references = []
sessions = ORTModelForSpeechSeq2Seq.load_model(
            os.path.join(model_path, 'encoder_model.onnx'),
            os.path.join(model_path, 'decoder_model.onnx'),
            os.path.join(model_path, 'decoder_with_past_model.onnx'))
model = ORTModelForSpeechSeq2Seq(sessions[0], sessions[1], model_config, model_path, sessions[2])
for idx, batch in enumerate(librispeech_test_clean):
    audio = batch["audio"]
    input_features = processor(audio["array"], sampling_rate=audio["sampling_rate"], return_tensors="pt").input_features
    reference = processor.tokenizer._normalize(batch['text'])
    references.append(reference)
    predicted_ids = model.generate(input_features)[0]
    transcription = processor.decode(predicted_ids)
    prediction = processor.tokenizer._normalize(transcription)
    predictions.append(prediction)
wer_result = wer.compute(references=references, predictions=predictions)
print(f"词错率结果: {wer_result * 100}")

性能指标: