app.py

app.py

@@ -0,0 +1,111 @@
+import pickle
+import json
+import torch
+import torch
+import torch.nn as nn
+import gradio as gr
+from torch.utils.data import Dataset
+from transformers import BertTokenizer, BertModel
+
+tokenizer = BertTokenizer.from_pretrained('bert-base-multilingual-cased')
+model = BertModel.from_pretrained('bert-base-multilingual-cased')
+device=torch.device("cuda" if torch.cuda.is_available() else "cpu")
+
+with open("word_to_index.pkl", "rb") as f:
+    word_to_index = pickle.load(f)
+
+# Loading index to word mapping
+with open("index_to_word.pkl", "rb") as f:
+    index_to_word = pickle.load(f)
+numclass=len(index_to_word)
+
+
+class DefinitionClassifier(nn.Module):
+    def __init__(self, model, tokenizer, num_classes):
+        super(DefinitionClassifier, self).__init__()
+        
+        # Load pre-trained Indic BERT model
+        self.bert = model
+        self.tokenizer = tokenizer
+        self.num_classes = num_classes
+        
+        # Define classification layer
+        self.classifier = nn.Linear(self.bert.config.hidden_size, num_classes)
+
+    def forward(self, input_ids, attention_mask):
+        # Forward pass through the BERT model
+        outputs = self.bert(input_ids, attention_mask)
+        
+        # Extract the CLS token embeddings
+        cls_embeddings = outputs.pooler_output
+        
+        # Pass the CLS embeddings through the classification layer
+        logits = self.classifier(cls_embeddings)
+        
+        return logits
+model_final = DefinitionClassifier(model=model,tokenizer=tokenizer,num_classes=numclass).to(device)
+state_dict = torch.load("/data5/home/shubhambhaws2/project_drona1/project/modelmbert.pth")
+
+# Load the state dictionary into the model
+model_final.load_state_dict(state_dict)
+
+def formate_story(story):
+    sen= story.split("\n")
+    c=0
+    sen[0]=sen[0]+"\n"
+    fin=[]
+    for i in sen:
+        if(c%2==0):
+            fin.append(i+"\n")
+        else:
+            fin.append(i)
+        c=c+1
+    return "".join(fin)
+
+print("Model Loaded")
+
+def generate_word(defination, k):
+    text = "the place where everyone can go"
+    inputs = tokenizer.encode_plus(
+        defination,
+        max_length=128,
+        padding='max_length',
+        truncation=True,
+        return_tensors='pt'
+    )
+    input_ids = inputs['input_ids'].to(device)
+    attention_mask = inputs['attention_mask'].to(device)
+    logits = model_final(input_ids, attention_mask)
+
+    probabilities = torch.softmax(logits, dim=1)
+    topk_probabilities, topk_indices = torch.topk(probabilities, k, dim=1)
+    pred = topk_indices.squeeze().cpu().numpy().tolist()
+    output = []
+
+    for i in pred:
+        output.append(index_to_word[i])
+    
+    # Return a string with each word on a new line
+    return "\n".join(output)
+
+def gradio_reset():
+    return None, None, 10
+
+# New Title
+title = """<h1 align='center'>Reverse Dictionary</a></h1>"""
+
+with gr.Blocks() as demo:
+    gr.Markdown(title)
+    story = gr.Textbox(label="Input Description", lines=2)
+    k = gr.Slider(label="Total Output", minimum=1, maximum=100, step=1, value=10)
+
+    with gr.Row():
+        upload_button = gr.Button(value="Generate Word", interactive=True, variant="primary")
+        clear = gr.Button("Clear")
+    output = gr.Textbox(label="Output Hindi word", lines=20)
+        
+    upload_button.click(generate_word, [story, k], [output])
+    clear.click(gradio_reset, [], [story, output,k])
+        
+demo.queue()
+demo.launch(share=True)