#!/usr/bin/env python3
from transformers import SpeechEncoderDecoderModel, AutoFeatureExtractor, AutoTokenizer
import torch


encoder_id = "facebook/wav2vec2-large-lv60"
decoder_id = "roberta-large"

model = SpeechEncoderDecoderModel.from_encoder_decoder_pretrained(encoder_id, decoder_id, encoder_add_adapter=False)
model.config.decoder_start_token_id = model.decoder.config.bos_token_id
model.config.pad_token_id = model.decoder.config.pad_token_id
model.config.eos_token_id = model.decoder.config.eos_token_id
model.config.max_length = 50
model.config.num_beams = 1
model.config.use_cache = False
model.config.decoder.use_cache = False
model.config.processor_class = "Wav2Vec2Processor"

# set all encoder regularisation to zero
model.config.encoder.feat_proj_dropout = 0.0
model.config.encoder.final_dropout = 0.0
model.config.encoder.activation_dropout = 0.0
model.config.encoder.apply_spec_augment = False
model.config.encoder.attention_dropout = 0.0
model.config.encoder.feat_extract_dropout = 0.0
model.config.encoder.feat_proj_dropout = 0.0
model.config.encoder.hidden_dropout = 0.0
model.config.encoder.hidden_dropout_prob = 0.0
model.config.encoder.layerdrop = 0.0
model.config.encoder.mask_feature_prob = 0.0
model.config.encoder.mask_time_prob = 0.0

# set all decoder regularisation to zero
model.config.decoder.attention_probs_dropout_prob = 0.0
model.config.decoder.hidden_dropout_prob = 0.0


# check if generation works
out = model.generate(torch.ones((1, 2000)))

model.save_pretrained("./")

feature_extractor = AutoFeatureExtractor.from_pretrained(encoder_id)
feature_extractor.save_pretrained("./")
tokenizer = AutoTokenizer.from_pretrained(decoder_id)
tokenizer.save_pretrained("./")