Upload config

audio_text_multimodal.py

@@ -0,0 +1,215 @@
+from dataclasses import dataclass
+from typing import Union, Type
+
+import torch
+from transformers.modeling_outputs import SequenceClassifierOutput
+from transformers import (
+    PreTrainedModel,
+    PretrainedConfig,
+    WavLMConfig,
+    BertConfig,
+    WavLMModel,
+    BertModel,
+    Wav2Vec2Config,
+    Wav2Vec2Model
+)
+
+
+class MultiModalConfig(PretrainedConfig):
+    """Base class for multimodal configs"""
+    def __init__(self, **kwargs):
+        super().__init__(**kwargs)
+
+
+class WavLMBertConfig(MultiModalConfig):
+    ...
+
+
+class BaseClassificationModel(PreTrainedModel):
+    config: Type[Union[PretrainedConfig, None]] = None
+
+    def compute_loss(self, logits, labels):
+        """Compute loss
+
+        Args:
+            logits (torch.FloatTensor): logits
+            labels (torch.LongTensor): labels
+
+        Returns:
+            torch.FloatTensor: loss
+
+        Raises:
+            ValueError: Invalid number of labels
+        """
+        if self.config.problem_type is None:
+            if self.num_labels == 1:
+                self.config.problem_type = "regression"
+            elif self.num_labels > 1:
+                self.config.problem_type = "single_label_classification"
+            else:
+                raise ValueError("Invalid number of labels: {}".format(self.num_labels))
+
+        if self.config.problem_type == "single_label_classification":
+            loss_fct = torch.nn.CrossEntropyLoss()
+            loss = loss_fct(logits.view(-1, self.num_labels), labels.view(-1))
+
+        elif self.config.problem_type == "multi_label_classification":
+            loss_fct = torch.nn.BCEWithLogitsLoss()
+            loss = loss_fct(logits.view(-1, self.num_labels), labels.view(-1, self.num_labels))
+
+        elif self.config.problem_type == "regression":
+            loss_fct = torch.nn.MSELoss()
+            loss = loss_fct(logits.view(-1), labels.view(-1))
+        else:
+            raise ValueError("Problem_type {} not supported".format(self.config.problem_type))
+
+        return loss
+
+    @staticmethod
+    def merged_strategy(
+            hidden_states,
+            mode="mean"
+    ):
+        """Merged strategy for pooling
+
+        Args:
+            hidden_states (torch.FloatTensor): hidden states
+            mode (str, optional): pooling mode. Defaults to "mean".
+
+        Returns:
+            torch.FloatTensor: pooled hidden states
+        """
+        if mode == "mean":
+            outputs = torch.mean(hidden_states, dim=1)
+        elif mode == "sum":
+            outputs = torch.sum(hidden_states, dim=1)
+        elif mode == "max":
+            outputs = torch.max(hidden_states, dim=1)[0]
+        else:
+            raise Exception(
+                "The pooling method hasn't been defined! Your pooling mode must be one of these ['mean', 'sum', 'max']")
+
+        return outputs
+
+
+class AudioTextModelForSequenceBaseClassification(BaseClassificationModel):
+    config_class = MultiModalConfig
+
+    def __init__(self, config):
+        """
+        Args:
+            config (MultiModalConfig): config
+
+        Attributes:
+            config (MultiModalConfig): config
+            num_labels (int): number of labels
+            audio_config (Union[PretrainedConfig, None]): audio config
+            text_config (Union[PretrainedConfig, None]): text config
+            audio_model (Union[PreTrainedModel, None]): audio model
+            text_model (Union[PreTrainedModel, None]): text model
+            classifier (Union[torch.nn.Linear, None]): classifier
+        """
+        super().__init__(config)
+        self.config = config
+        self.num_labels = self.config.num_labels
+        self.audio_config: Union[PretrainedConfig, None] = None
+        self.text_config: Union[PretrainedConfig, None] = None
+        self.audio_model: Union[PreTrainedModel, None] = None
+        self.text_model: Union[PreTrainedModel, None] = None
+        self.classifier: Union[torch.nn.Linear, None] = None
+
+    def forward(
+            self,
+            input_ids=None,
+            input_values=None,
+            text_attention_mask=None,
+            audio_attention_mask=None,
+            token_type_ids=None,
+            position_ids=None,
+            head_mask=None,
+            inputs_embeds=None,
+            labels=None,
+            output_attentions=None,
+            output_hidden_states=None,
+            return_dict=True,
+    ):
+        """Forward method for multimodal model for sequence classification task (e.g. text + audio)
+
+        Args:
+            input_ids (torch.LongTensor, optional): input ids. Defaults to None.
+            input_values (torch.FloatTensor, optional): input values. Defaults to None.
+            text_attention_mask (torch.LongTensor, optional): text attention mask. Defaults to None.
+            audio_attention_mask (torch.LongTensor, optional): audio attention mask. Defaults to None.
+            token_type_ids (torch.LongTensor, optional): token type ids. Defaults to None.
+            position_ids (torch.LongTensor, optional): position ids. Defaults to None.
+            head_mask (torch.FloatTensor, optional): head mask. Defaults to None.
+            inputs_embeds (torch.FloatTensor, optional): inputs embeds. Defaults to None.
+            labels (torch.LongTensor, optional): labels. Defaults to None.
+            output_attentions (bool, optional): output attentions. Defaults to None.
+            output_hidden_states (bool, optional): output hidden states. Defaults to None.
+            return_dict (bool, optional): return dict. Defaults to True.
+
+        Returns:
+            torch.FloatTensor: logits
+        """
+        audio_output = self.audio_model(
+            input_values=input_values,
+            attention_mask=audio_attention_mask,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict
+        )
+        text_output = self.text_model(
+            input_ids=input_ids,
+            attention_mask=text_attention_mask,
+            token_type_ids=token_type_ids,
+            position_ids=position_ids,
+            head_mask=head_mask,
+            inputs_embeds=inputs_embeds,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+        audio_mean = self.merged_strategy(audio_output.last_hidden_state, mode="mean")
+
+        pooled_output = torch.cat(
+            (audio_mean, text_output.pooler_output), dim=1
+        )
+        logits = self.classifier(pooled_output)
+        loss = None
+
+        if labels is not None:
+            loss = self.compute_loss(logits, labels)
+
+        return SequenceClassifierOutput(
+            loss=loss,
+            logits=logits
+        )
+
+
+class WavLMBertForSequenceClassification(AudioTextModelForSequenceBaseClassification):
+    """
+    WavLMBertForSequenceClassification is a model for sequence classification task
+     (e.g. sentiment analysis, text classification, etc.)
+
+    Args:
+        config (WavLMBertConfig): config
+
+    Attributes:
+        config (WavLMBertConfig): config
+        audio_config (WavLMConfig): wav2vec2 config
+        text_config (BertConfig): bert config
+        audio_model (WavLMModel): wav2vec2 model
+        text_model (BertModel): bert model
+        classifier (torch.nn.Linear): classifier
+    """
+    def __init__(self, config):
+        super().__init__(config)
+        self.audio_config = WavLMConfig.from_dict(self.config.WavLMModel)
+        self.text_config = BertConfig.from_dict(self.config.BertModel)
+        self.audio_model = WavLMModel(self.audio_config)
+        self.text_model = BertModel(self.text_config)
+        self.classifier = torch.nn.Linear(
+            self.audio_config.hidden_size + self.text_config.hidden_size, self.num_labels
+        )
+        self.init_weights()

config.json

@@ -0,0 +1,293 @@
+{
+  "BertModel": {
+    "_name_or_path": "DeepPavlov/rubert-base-cased",
+    "add_cross_attention": false,
+    "architectures": [
+      "BertModel"
+    ],
+    "attention_probs_dropout_prob": 0.1,
+    "bad_words_ids": null,
+    "begin_suppress_tokens": null,
+    "bos_token_id": null,
+    "chunk_size_feed_forward": 0,
+    "classifier_dropout": null,
+    "cross_attention_hidden_size": null,
+    "decoder_start_token_id": null,
+    "directionality": "bidi",
+    "diversity_penalty": 0.0,
+    "do_sample": false,
+    "early_stopping": false,
+    "encoder_no_repeat_ngram_size": 0,
+    "eos_token_id": null,
+    "exponential_decay_length_penalty": null,
+    "finetuning_task": null,
+    "forced_bos_token_id": null,
+    "forced_eos_token_id": null,
+    "hidden_act": "gelu",
+    "hidden_dropout_prob": 0.1,
+    "hidden_size": 768,
+    "id2label": {
+      "0": "LABEL_0",
+      "1": "LABEL_1"
+    },
+    "initializer_range": 0.02,
+    "intermediate_size": 3072,
+    "is_decoder": false,
+    "is_encoder_decoder": false,
+    "label2id": {
+      "LABEL_0": 0,
+      "LABEL_1": 1
+    },
+    "layer_norm_eps": 1e-12,
+    "length_penalty": 1.0,
+    "max_length": 20,
+    "max_position_embeddings": 512,
+    "min_length": 0,
+    "model_type": "bert",
+    "no_repeat_ngram_size": 0,
+    "num_attention_heads": 12,
+    "num_beam_groups": 1,
+    "num_beams": 1,
+    "num_hidden_layers": 12,
+    "num_return_sequences": 1,
+    "output_attentions": false,
+    "output_hidden_states": false,
+    "output_past": true,
+    "output_scores": false,
+    "pad_token_id": 0,
+    "pooler_fc_size": 768,
+    "pooler_num_attention_heads": 12,
+    "pooler_num_fc_layers": 3,
+    "pooler_size_per_head": 128,
+    "pooler_type": "first_token_transform",
+    "position_embedding_type": "absolute",
+    "prefix": null,
+    "problem_type": null,
+    "pruned_heads": {},
+    "remove_invalid_values": false,
+    "repetition_penalty": 1.0,
+    "return_dict": true,
+    "return_dict_in_generate": false,
+    "sep_token_id": null,
+    "suppress_tokens": null,
+    "task_specific_params": null,
+    "temperature": 1.0,
+    "tf_legacy_loss": false,
+    "tie_encoder_decoder": false,
+    "tie_word_embeddings": true,
+    "tokenizer_class": null,
+    "top_k": 50,
+    "top_p": 1.0,
+    "torch_dtype": null,
+    "torchscript": false,
+    "transformers_version": "4.27.4",
+    "type_vocab_size": 2,
+    "typical_p": 1.0,
+    "use_bfloat16": false,
+    "use_cache": true,
+    "vocab_size": 119547
+  },
+  "WavLMModel": {
+    "_name_or_path": "jonatasgrosman/exp_w2v2t_ru_wavlm_s363",
+    "activation_dropout": 0.05,
+    "adapter_kernel_size": 3,
+    "adapter_stride": 2,
+    "add_adapter": false,
+    "add_cross_attention": false,
+    "apply_spec_augment": true,
+    "architectures": [
+      "WavLMModel"
+    ],
+    "attention_dropout": 0.05,
+    "bad_words_ids": null,
+    "begin_suppress_tokens": null,
+    "bos_token_id": 1,
+    "chunk_size_feed_forward": 0,
+    "classifier_proj_size": 256,
+    "codevector_dim": 768,
+    "contrastive_logits_temperature": 0.1,
+    "conv_bias": false,
+    "conv_dim": [
+      512,
+      512,
+      512,
+      512,
+      512,
+      512,
+      512
+    ],
+    "conv_kernel": [
+      10,
+      3,
+      3,
+      3,
+      3,
+      2,
+      2
+    ],
+    "conv_stride": [
+      5,
+      2,
+      2,
+      2,
+      2,
+      2,
+      2
+    ],
+    "cross_attention_hidden_size": null,
+    "ctc_loss_reduction": "sum",
+    "ctc_zero_infinity": false,
+    "decoder_start_token_id": null,
+    "diversity_loss_weight": 0.1,
+    "diversity_penalty": 0.0,
+    "do_sample": false,
+    "do_stable_layer_norm": true,
+    "early_stopping": false,
+    "encoder_no_repeat_ngram_size": 0,
+    "eos_token_id": 2,
+    "exponential_decay_length_penalty": null,
+    "feat_extract_activation": "gelu",
+    "feat_extract_dropout": 0.0,
+    "feat_extract_norm": "layer",
+    "feat_proj_dropout": 0.05,
+    "feat_quantizer_dropout": 0.0,
+    "final_dropout": 0.05,
+    "finetuning_task": null,
+    "forced_bos_token_id": null,
+    "forced_eos_token_id": null,
+    "gradient_checkpointing": false,
+    "hidden_act": "gelu",
+    "hidden_dropout": 0.05,
+    "hidden_size": 1024,
+    "id2label": {
+      "0": "LABEL_0",
+      "1": "LABEL_1"
+    },
+    "initializer_range": 0.02,
+    "intermediate_size": 4096,
+    "is_decoder": false,
+    "is_encoder_decoder": false,
+    "label2id": {
+      "LABEL_0": 0,
+      "LABEL_1": 1
+    },
+    "layer_norm_eps": 1e-05,
+    "layerdrop": 0.05,
+    "length_penalty": 1.0,
+    "mask_channel_length": 10,
+    "mask_channel_min_space": 1,
+    "mask_channel_other": 0.0,
+    "mask_channel_prob": 0.0,
+    "mask_channel_selection": "static",
+    "mask_feature_length": 10,
+    "mask_feature_min_masks": 0,
+    "mask_feature_prob": 0.0,
+    "mask_time_length": 10,
+    "mask_time_min_masks": 2,
+    "mask_time_min_space": 1,
+    "mask_time_other": 0.0,
+    "mask_time_prob": 0.05,
+    "mask_time_selection": "static",
+    "max_bucket_distance": 800,
+    "max_length": 20,
+    "min_length": 0,
+    "model_type": "wavlm",
+    "no_repeat_ngram_size": 0,
+    "num_adapter_layers": 3,
+    "num_attention_heads": 16,
+    "num_beam_groups": 1,
+    "num_beams": 1,
+    "num_buckets": 320,
+    "num_codevector_groups": 2,
+    "num_codevectors_per_group": 320,
+    "num_conv_pos_embedding_groups": 16,
+    "num_conv_pos_embeddings": 128,
+    "num_ctc_classes": 80,
+    "num_feat_extract_layers": 7,
+    "num_hidden_layers": 24,
+    "num_negatives": 100,
+    "num_return_sequences": 1,
+    "output_attentions": false,
+    "output_hidden_size": 1024,
+    "output_hidden_states": false,
+    "output_scores": false,
+    "pad_token_id": 0,
+    "prefix": null,
+    "problem_type": null,
+    "proj_codevector_dim": 768,
+    "pruned_heads": {},
+    "remove_invalid_values": false,
+    "repetition_penalty": 1.0,
+    "replace_prob": 0.5,
+    "return_dict": true,
+    "return_dict_in_generate": false,
+    "sep_token_id": null,
+    "suppress_tokens": null,
+    "task_specific_params": null,
+    "tdnn_dilation": [
+      1,
+      2,
+      3,
+      1,
+      1
+    ],
+    "tdnn_dim": [
+      512,
+      512,
+      512,
+      512,
+      1500
+    ],
+    "tdnn_kernel": [
+      5,
+      3,
+      3,
+      1,
+      1
+    ],
+    "temperature": 1.0,
+    "tf_legacy_loss": false,
+    "tie_encoder_decoder": false,
+    "tie_word_embeddings": true,
+    "tokenizer_class": "Wav2Vec2CTCTokenizer",
+    "top_k": 50,
+    "top_p": 1.0,
+    "torch_dtype": "float32",
+    "torchscript": false,
+    "transformers_version": "4.27.4",
+    "typical_p": 1.0,
+    "use_bfloat16": false,
+    "use_weighted_layer_sum": false,
+    "vocab_size": 40,
+    "xvector_output_dim": 512
+  },
+  "_name_or_path": "Ar4ikov/wavlm-bert-base-multimodal-emotion-russian-resd",
+  "architectures": [
+    "WavLMBertForSequenceClassification"
+  ],
+  "auto_map": {
+    "AutoConfig": "audio_text_multimodal.WavLMBertConfig",
+    "AutoModel": "audio_text_multimodal.WavLMBertForSequenceClassification"
+  },
+  "id2label": {
+    "0": "anger",
+    "1": "disgust",
+    "2": "enthusiasm",
+    "3": "fear",
+    "4": "happiness",
+    "5": "neutral",
+    "6": "sadness"
+  },
+  "label2id": {
+    "anger": 0,
+    "disgust": 1,
+    "enthusiasm": 2,
+    "fear": 3,
+    "happiness": 4,
+    "neutral": 5,
+    "sadness": 6
+  },
+  "problem_type": "single_label_classification",
+  "torch_dtype": "float32",
+  "transformers_version": "4.27.4"
+}