add: translation codes for hf space

app.py

@@ -0,0 +1,40 @@
+import gradio as gr
+from translator import GlossTranslator
+
+# Configuration
+BASE_MODEL_ID: str = "facebook/mbart-large-50"
+PEFT_MODEL_ID: str = "ayhay/BanglaText2Gloss"
+
+# Initialize translator instance
+translator_engine: GlossTranslator = GlossTranslator(
+    base_model_id=BASE_MODEL_ID, peft_model_id=PEFT_MODEL_ID
+)
+
+
+def process_translation(input_text: str) -> str:
+    """Wrapper function to handle UI interactions and catch errors safely."""
+    try:
+        output: str = translator_engine.translate(input_text)
+        return output
+    except Exception as e:
+        return f"System Error: {str(e)}"
+
+
+def create_ui() -> gr.Interface:
+    """Builds and returns the Gradio interface."""
+    interface: gr.Interface = gr.Interface(
+        fn=process_translation,
+        inputs=gr.Textbox(
+            lines=3, placeholder="Type Bangla text here...", label="Input: Bangla"
+        ),
+        outputs=gr.Textbox(label="Output: Sign Language Gloss"),
+        title="Bangla to Sign Language Gloss",
+        description="A specialized translation model based on mBART-large-50.",
+        examples=[["আপনি কেমন আছেন?"]],
+    )
+    return interface
+
+
+if __name__ == "__main__":
+    app: gr.Interface = create_ui()
+    app.launch()

requirements.txt

@@ -0,0 +1,5 @@
+transformers
+peft
+torch
+gradio
+sentencepiece

translator.py

@@ -0,0 +1,38 @@
+import torch
+from transformers import (
+    MBartForConditionalGeneration, 
+    AutoTokenizer, 
+    PreTrainedTokenizer, 
+    PreTrainedTokenizerFast
+)
+from peft import PeftModel
+from typing import Union, Dict, List
+
+TokenizerType = Union[PreTrainedTokenizer, PreTrainedTokenizerFast]
+
+class GlossTranslator:
+    def __init__(self, base_model_id: str, peft_model_id: str) -> None:
+        self.base_model_id: str = base_model_id
+        self.peft_model_id: str = peft_model_id
+        
+        # Load Tokenizer
+        self.tokenizer: TokenizerType = AutoTokenizer.from_pretrained(self.peft_model_id)
+        
+        # Load Base Model and apply PEFT adapters
+        base_model: MBartForConditionalGeneration = MBartForConditionalGeneration.from_pretrained(self.base_model_id)
+        self.model: PeftModel = PeftModel.from_pretrained(base_model, self.peft_model_id)
+        
+        # Optimize for inference
+        self.model.eval()
+
+    def translate(self, text: str) -> str:
+        if not text.strip():
+            return ""
+            
+        inputs: Dict[str, torch.Tensor] = self.tokenizer(text, return_tensors="pt")
+        
+        with torch.no_grad():
+            output_tokens: torch.Tensor = self.model.generate(**inputs, max_new_tokens=50)
+            
+        decoded_glosses: List[str] = self.tokenizer.batch_decode(output_tokens, skip_special_tokens=True)
+        return decoded_glosses[0]