Jingya/wav2vec2-large-960h-lv60-self-neuronx-audio-classification

This is model is compiled explicitly for AWS Neuronx(inferentia 2 / trainium 1) with the following codes:

from datasets import load_dataset
from transformers import AutoProcessor

from optimum.neuron import NeuronModelForAudioClassification, pipeline


dataset = load_dataset("hf-internal-testing/librispeech_asr_demo", "clean", split="validation")
dataset = dataset.sort("id")
sampling_rate = dataset.features["audio"].sampling_rate

model_id = "facebook/wav2vec2-large-960h-lv60-self"
processor = AutoProcessor.from_pretrained(model_id)
input_shapes = {"batch_size": 1, "audio_sequence_length": 100000}
compiler_args = {"auto_cast": "matmul", "auto_cast_type": "bf16"}
model = NeuronModelForAudioClassification.from_pretrained(
    model_id,
    export=True,
    disable_neuron_cache=True,
    **input_shapes,
    **compiler_args,
)
model.save_pretrained("wav2vec2_neuron")