Initial commit

app.py

@@ -0,0 +1,62 @@
+import gradio as gr
+import torch
+from dual_regression_model import DualRegressionModel
+import transformers
+from transformers import pipeline
+from functools import partial
+
+# load the models
+# CLF: A-pt-bs16-dbmdz-bert-base-italian-cased
+clf_model_tag = "clf_model/"
+clf_tokenizer = transformers.AutoTokenizer.from_pretrained(clf_model_tag)
+clf_model = transformers.AutoModelForSequenceClassification.from_pretrained(clf_model_tag)
+clf_pipeline = pipeline("text-classification", model=clf_model, tokenizer=clf_tokenizer)
+
+# REG
+reg_model_tag = "distilbert-base-multilingual-cased"
+reg_model_folder = "reg_model/regression_model.pt"
+reg_model = DualRegressionModel(model_name_or_path=reg_model_tag)
+reg_model.load_model(reg_model_folder)
+
+
+# define the function to be used for prediction
+def predict(text):
+    # predict the class
+    clf_prediction = clf_pipeline(text)[0]
+    # predict the coordinates
+    reg_input = reg_model.tokenizer(text, return_tensors="pt")
+    reg_prediction = reg_model(reg_input)
+    latitude, longitude = reg_prediction["latitude"].item(), reg_prediction["longitude"].item()
+    lat_min = 38
+    lat_max = 46
+    long_min = 8
+    long_max = 18
+    # return the results
+    html_output = f"<h3>The identified region is: {clf_prediction['label']}</h3>"
+    # plot points on the map of Italy
+    html_output += f'<h3>Predicted point on map:</h3><p>Latitude: {latitude}</p><p>Longitude: {longitude}</p>'
+    html_output += f'<iframe width="425" height="350" frameborder="0" scrolling="no" marginheight="0" marginwidth="0" src="https://www.openstreetmap.org/export/embed.html?bbox={long_min}%2C{lat_min}%2C{long_max}%2C{lat_max}&amp;layer=mapnik&marker={latitude}%2C{longitude}" style="border: 1px solid black"></iframe><br/><small><a href="https://www.openstreetmap.org/#map=13/{latitude}/{longitude}">Visualizza mappa ingrandita</a></small>'
+
+    return html_output
+
+# --------------------------------------------------------------------------------------------
+# Gradio interface
+# --------------------------------------------------------------------------------------------
+
+# define the interface
+iface = gr.Interface(
+    fn=predict,
+    inputs=gr.Textbox(lines=2, placeholder="Insert the text here..."),
+    outputs=gr.HTML(),
+    title="DANTE: Dialect ANalysis TEam",
+    description="This is a demo of a classification and regression model for locating the italian dialect of a given text.",
+    examples=[
+        ["Bisognerebbe saperli materializzare .... !!  Ma  ovviamente  .. belin .... NO SE PEU SCIUSCIA' E SCIORBI'"],
+        ["Guaglio' Buongiorno! Azz! Vir te si scurdat puparuol e mulignane pero '!! E che se fa😑"],
+        ["Il massimo...ghe ne minga par nisun"],
+        ["Che poi a me la tuta piace na cifra da vede. Subisco un po' lo stigma sociale che noi con la fregna dovemo stà sempre apposto.",]
+    ]
+)
+
+# launch the interface
+iface.launch()

clf_model/.gitattributes

@@ -0,0 +1,34 @@
+*.7z filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.bz2 filter=lfs diff=lfs merge=lfs -text
+*.ckpt filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.gz filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.mlmodel filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.npy filter=lfs diff=lfs merge=lfs -text
+*.npz filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.parquet filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pickle filter=lfs diff=lfs merge=lfs -text
+*.pkl filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*.rar filter=lfs diff=lfs merge=lfs -text
+*.safetensors filter=lfs diff=lfs merge=lfs -text
+saved_model/**/* filter=lfs diff=lfs merge=lfs -text
+*.tar.* filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tgz filter=lfs diff=lfs merge=lfs -text
+*.wasm filter=lfs diff=lfs merge=lfs -text
+*.xz filter=lfs diff=lfs merge=lfs -text
+*.zip filter=lfs diff=lfs merge=lfs -text
+*.zst filter=lfs diff=lfs merge=lfs -text
+*tfevents* filter=lfs diff=lfs merge=lfs -text

clf_model/config.json

@@ -0,0 +1,71 @@
+{
+  "_name_or_path": "best_ft_models_a/PT/16BS/dbmdz-bert-base-italian-cased/best_model/",
+  "architectures": [
+    "BertForSequenceClassification"
+  ],
+  "attention_probs_dropout_prob": 0.1,
+  "classifier_dropout": null,
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 768,
+  "id2label": {
+    "0": "Abruzzo",
+    "1": "Basilicata",
+    "2": "Calabria",
+    "3": "Campania",
+    "4": "Emilia Romagna",
+    "5": "Friuli-Venezia Giulia",
+    "6": "Lazio",
+    "7": "Liguria",
+    "8": "Lombardia",
+    "9": "Marche",
+    "10": "Molise",
+    "11": "Piemonte",
+    "12": "Puglia",
+    "13": "Sardegna",
+    "14": "Sicilia",
+    "15": "Toscana",
+    "16": "Trentino-Alto Adige",
+    "17": "Umbria",
+    "18": "Valle d'Aosta",
+    "19": "Veneto"
+  },
+  "initializer_range": 0.02,
+  "intermediate_size": 3072,
+  "label2id": {
+    "Abruzzo": 0,
+    "Basilicata": 1,
+    "Calabria": 2,
+    "Campania": 3,
+    "Emilia Romagna": 4,
+    "Friuli-Venezia Giulia": 5,
+    "Lazio": 6,
+    "Liguria": 7,
+    "Lombardia": 8,
+    "Marche": 9,
+    "Molise": 10,
+    "Piemonte": 11,
+    "Puglia": 12,
+    "Sardegna": 13,
+    "Sicilia": 14,
+    "Toscana": 15,
+    "Trentino-Alto Adige": 16,
+    "Umbria": 17,
+    "Valle d'Aosta": 18,
+    "Veneto": 19
+  },
+  "layer_norm_eps": 1e-12,
+  "max_position_embeddings": 512,
+  "model_type": "bert",
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "output_hidden_states": true,
+  "pad_token_id": 0,
+  "position_embedding_type": "absolute",
+  "problem_type": "single_label_classification",
+  "torch_dtype": "float32",
+  "transformers_version": "4.26.1",
+  "type_vocab_size": 2,
+  "use_cache": true,
+  "vocab_size": 31102
+}

clf_model/pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7b9f16887a3e26030e3d57905976b8eb36a6e5846b1047fbabde05b62adc605b
+size 439842229

clf_model/special_tokens_map.json

@@ -0,0 +1,7 @@
+{
+  "cls_token": "[CLS]",
+  "mask_token": "[MASK]",
+  "pad_token": "[PAD]",
+  "sep_token": "[SEP]",
+  "unk_token": "[UNK]"
+}

clf_model/tokenizer_config.json

@@ -0,0 +1,17 @@
+{
+  "cls_token": "[CLS]",
+  "do_basic_tokenize": true,
+  "do_lower_case": false,
+  "mask_token": "[MASK]",
+  "max_len": 512,
+  "model_max_length": 512,
+  "name_or_path": "best_ft_models_a/PT/16BS/dbmdz-bert-base-italian-cased/best_model/",
+  "never_split": null,
+  "pad_token": "[PAD]",
+  "sep_token": "[SEP]",
+  "special_tokens_map_file": null,
+  "strip_accents": null,
+  "tokenize_chinese_chars": true,
+  "tokenizer_class": "BertTokenizer",
+  "unk_token": "[UNK]"
+}

dual_regression_model.py

@@ -0,0 +1,94 @@
+import numpy as np
+import torch
+from torch import nn
+from transformers import AutoModel, AutoModelForMaskedLM, AutoTokenizer
+
+
+class DualRegressionModel(nn.Module):
+    def __init__(
+        self,
+        model_name_or_path: str = "camembert/camembert-base",
+        loss_aggreatation: str = "mean",
+    ):
+        """
+        This class instantiates the pre-training model.
+        :param model_name_or_path: The name or path of the model to be used for pre-training.
+        """
+
+        super().__init__()
+        if "bart" in model_name_or_path:
+            self.model = AutoModel.from_pretrained(
+                model_name_or_path, output_hidden_states=True
+            )
+            self.model = self.model.encoder
+        else:
+            self.model = AutoModelForMaskedLM.from_pretrained(
+                model_name_or_path, output_hidden_states=True
+            )
+
+        self.tokenizer = AutoTokenizer.from_pretrained(model_name_or_path)
+        self.loss_aggreatation = loss_aggreatation
+
+        # create two different regression heads for two tasks (latitude and longitude)
+        self.lat_regression_head = torch.nn.Linear(self.model.config.hidden_size, 1)
+        self.long_regression_head = torch.nn.Linear(self.model.config.hidden_size, 1)
+
+        self.crierion = torch.nn.MSELoss()
+
+    def forward(
+        self,
+        batch,
+    ):
+        """
+        This function is called to compute the loss for the specified task.
+        :param batch: The batch of data.
+        """
+        predict = not batch.keys() & {"longitude", "latitude"}
+
+        input_ids = batch["input_ids"]
+        attention_mask = batch["attention_mask"]
+        if not predict:
+            latitudes = batch["latitude"]
+            longitudes = batch["longitude"]
+
+        # get the last hidden state
+        last_hidden_state = self.model(
+            input_ids=input_ids,
+            attention_mask=attention_mask,
+        ).hidden_states[-1][:, 0, :]
+
+        lat_predictions = self.lat_regression_head(last_hidden_state)
+        long_predictions = self.long_regression_head(last_hidden_state)
+
+        result = {"latitude": lat_predictions, "longitude": long_predictions}
+
+        if not predict:
+            lat_loss = self.crierion(lat_predictions.squeeze(), latitudes)
+            long_loss = self.crierion(long_predictions.squeeze(), longitudes)
+
+            if self.loss_aggreatation == "mean":
+                loss = (lat_loss + long_loss) / 2
+            elif self.loss_aggreatation == "sum":
+                loss = lat_loss + long_loss
+            else:
+                raise ValueError("Only mean and sum are supported for loss aggregation")
+            result |= {"loss": loss}
+
+        return result
+
+    def save_model(self, path):
+        """
+        This function is called to save the model to a specified path. E.g. "model.pt"
+        :param path: The path where the model is saved.
+        """
+
+        torch.save(self.state_dict(), path)
+
+    def load_model(self, path):
+        """
+        This function is called to load the model.
+        :param path: The path where the model is saved. E.g. "model.pt"
+        """
+
+        # load the state dict
+        self.load_state_dict(torch.load(path))

reg_model/regression_model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2e477574615af95cc93b7bcb8173d7c000cf6a5c2d172d7ad4557a3b396652cd
+size 541834318

reg_model/special_tokens_map.json

@@ -0,0 +1,7 @@
+{
+  "cls_token": "[CLS]",
+  "mask_token": "[MASK]",
+  "pad_token": "[PAD]",
+  "sep_token": "[SEP]",
+  "unk_token": "[UNK]"
+}

reg_model/tokenizer_config.json

@@ -0,0 +1,14 @@
+{
+  "cls_token": "[CLS]",
+  "do_lower_case": false,
+  "mask_token": "[MASK]",
+  "model_max_length": 512,
+  "name_or_path": "DGMS/distilbert-base-multilingual-cased-dialect",
+  "pad_token": "[PAD]",
+  "sep_token": "[SEP]",
+  "special_tokens_map_file": null,
+  "strip_accents": null,
+  "tokenize_chinese_chars": true,
+  "tokenizer_class": "DistilBertTokenizer",
+  "unk_token": "[UNK]"
+}

requirements.txt

@@ -0,0 +1,4 @@
+torch
+transformers
+gradio
+numpy