import os 
import sys 
os.system("pip install transformers==4.27.0")
os.system("pip install torch")
os.system("pip install openai")
os.system("pip install accelerate")
from transformers import pipeline, WhisperModel, WhisperTokenizer, WhisperFeatureExtractor
os.system("pip install evaluate")
#import evaluate
#os.system("pip install evaluate[evaluator]")
os.system("pip install datasets")
# os.system("pip install llvmlite")
os.system("pip install spicy==1.8.1")
os.system("pip install soundfile")
os.system("pip install jiwer")
os.system("pip install datasets[audio]")
os.system("pip install numba==0.51.2")
from evaluate import evaluator
from datasets import load_dataset, Audio, disable_caching, set_caching_enabled

set_caching_enabled(False)
disable_caching()

huggingface_token = os.environ["huggingface_token"]


model = WhisperModel.from_pretrained("mskov/whisper_miso", use_auth_token=huggingface_token)
feature_extractor = AutoFeatureExtractor.from_pretrained("mskov/whisper_miso", use_auth_token=huggingface_token)
ds = load_dataset("mskov.ESC50", split="test")
inputs = feature_extractor(ds[0]["audio"]["array"], return_tensors="pt")
input_features = inputs.input_features
decoder_input_ids = torch.tensor([[1, 1]]) * model.config.decoder_start_token_id
last_hidden_state = model(input_features, decoder_input_ids=decoder_input_ids).last_hidden_state
list(last_hidden_state.shape)
print(list(last_hidden_state.shape))