Update app1.py

app1.py

@@ -0,0 +1,328 @@
+import os
+import sys
+import site
+import torch
+import transformers
+import gradio as gr
+import re
+
+from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
+
+print("=" * 80)
+print("Vidyaapati Translator")
+print("=" * 80)
+
+print("Python :", sys.version)
+print("Transformers :", transformers.__version__)
+
+# -----------------------------------------------------------------------------
+# Fix IndicTransToolkit import
+# -----------------------------------------------------------------------------
+
+def fix_collator():
+    try:
+        collator_path = None
+
+        for p in site.getsitepackages():
+            test = os.path.join(
+                p,
+                "IndicTransToolkit",
+                "collator.py"
+            )
+
+            if os.path.exists(test):
+                collator_path = test
+                break
+
+        if collator_path:
+
+            with open(collator_path, "r", encoding="utf-8") as f:
+                content = f.read()
+
+            content = content.replace(
+                "from transformers.tokenization_utils import PreTrainedTokenizerBase",
+                "from transformers import PreTrainedTokenizerBase",
+            )
+
+            with open(collator_path, "w", encoding="utf-8") as f:
+                f.write(content)
+
+            print("✓ IndicTransToolkit fixed")
+
+    except Exception as e:
+        print(e)
+
+
+fix_collator()
+
+from IndicTransToolkit import IndicProcessor
+
+DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
+
+print("Using device:", DEVICE)
+
+if DEVICE == "cuda":
+    print(torch.cuda.get_device_name(0))
+
+# -----------------------------------------------------------------------------
+# Model IDs
+# -----------------------------------------------------------------------------
+
+HI_KO_MODEL = "shriyadiker/KonkaniTrans-Hi-Ko"
+KO_HI_MODEL = "shriyadiker/KonkaniTrans-Ko-Hi"
+
+print("\nLoading Hindi → Konkani model...")
+
+hi_ko_tokenizer = AutoTokenizer.from_pretrained(
+    HI_KO_MODEL,
+    trust_remote_code=True,
+)
+
+hi_ko_model = AutoModelForSeq2SeqLM.from_pretrained(
+    HI_KO_MODEL,
+    trust_remote_code=True,
+    torch_dtype=torch.float16 if DEVICE == "cuda" else torch.float32,
+).to(DEVICE)
+
+print("✓ Hindi → Konkani loaded")
+
+print("\nLoading Konkani → Hindi model...")
+
+ko_hi_tokenizer = AutoTokenizer.from_pretrained(
+    KO_HI_MODEL,
+    trust_remote_code=True,
+)
+
+ko_hi_model = AutoModelForSeq2SeqLM.from_pretrained(
+    KO_HI_MODEL,
+    trust_remote_code=True,
+    torch_dtype=torch.float16 if DEVICE == "cuda" else torch.float32,
+).to(DEVICE)
+
+print("✓ Konkani → Hindi loaded")
+
+ip = IndicProcessor(inference=True)
+
+print("\nAll models loaded successfully.")
+
+# -----------------------------------------------------------------------------
+# Translation
+# -----------------------------------------------------------------------------
+
+def translate(text, direction):
+
+    if text is None or text.strip() == "":
+        return ""
+
+    try:
+
+        if direction == "Hindi → Konkani":
+
+            tokenizer = hi_ko_tokenizer
+            model = hi_ko_model
+
+            src_lang = "hin_Deva"
+            tgt_lang = "gom_Deva"
+
+        else:
+
+            tokenizer = ko_hi_tokenizer
+            model = ko_hi_model
+
+            src_lang = "gom_Deva"
+            tgt_lang = "hin_Deva"
+
+        batch = ip.preprocess_batch(
+            [text],
+            src_lang=src_lang,
+            tgt_lang=tgt_lang,
+        )
+
+        inputs = tokenizer(
+            batch,
+            truncation=True,
+            padding=True,
+            return_tensors="pt",
+            return_attention_mask=True,
+        ).to(DEVICE)
+
+        with torch.no_grad():
+
+            generated = model.generate(
+                **inputs,
+                max_length=256,
+                num_beams=5,
+                early_stopping=True,
+            )
+
+        decoded = tokenizer.batch_decode(
+            generated,
+            skip_special_tokens=True,
+            clean_up_tokenization_spaces=True,
+        )
+
+        output = ip.postprocess_batch(
+            decoded,
+            lang=tgt_lang,
+        )[0]
+
+        output = fix_escaped_unicode(output)
+
+        return output
+
+    except Exception as e:
+
+        return str(e)
+
+
+UNICODE_ESCAPE_PATTERN = re.compile(r'\\u([0-9a-fA-F]{4})')
+
+def fix_escaped_unicode(text):
+
+    if not isinstance(text, str):
+        return text
+
+    if "\\u" not in text:
+        return text
+
+    text = UNICODE_ESCAPE_PATTERN.sub(
+        lambda m: chr(int(m.group(1), 16)),
+        text
+    )
+
+    text = text.replace("ऱ", "ऱ")
+    text = text.replace("ऴ", "ऴ")
+
+    return text
+
+# -----------------------------------------------------------------------------
+# CSS
+# -----------------------------------------------------------------------------
+
+css = """
+footer {
+    display: none !important;
+}
+
+.gradio-container {
+    max-width: 800px !important;
+    margin: auto !important;
+    padding: 30px 20px !important;
+}
+
+textarea {
+    font-size: 18px !important;
+    font-family: 'Segoe UI', 'Noto Sans Devanagari', Arial, sans-serif !important;
+    line-height: 1.8 !important;
+    padding: 15px !important;
+    min-height: 180px !important;
+}
+
+button {
+    padding: 12px 40px !important;
+    font-size: 16px !important;
+    font-weight: 600 !important;
+    border-radius: 8px !important;
+}
+
+.title {
+    text-align: center;
+    margin-bottom: 30px;
+}
+
+.title h1 {
+    font-size: 28px;
+    font-weight: 600;
+    color: #1a1a1a;
+    margin-bottom: 5px;
+}
+
+.title p {
+    color: #666;
+    font-size: 14px;
+}
+"""
+
+# -----------------------------------------------------------------------------
+# Helper Functions
+# -----------------------------------------------------------------------------
+
+def clear_text():
+    return "", ""
+
+# -----------------------------------------------------------------------------
+# Gradio UI - Very Simple
+# -----------------------------------------------------------------------------
+
+with gr.Blocks(
+    title="VIDYAAPATI (Hindi-Konkani) | Goa University",
+    css=css,
+) as demo:
+
+    # Updated Heading with Project Description & Affiliation
+    gr.HTML("""
+    <div style="text-align: center; margin-bottom: 25px;">
+        <h2 style="margin-bottom: 6px; color: #1a1a1a; font-size: 24px;">VIDYAAPATI (Hindi-Konkani)</h2>
+        <div style="font-weight: 600; color: #4a5568; margin-bottom: 10px;">Goa University</div>
+        <p style="font-size: 13px; color: #666; line-height: 1.5; max-width: 750px; margin: 0 auto;">
+            <b>Project:</b> “VIDYAAPATI: Bidirectional Machine Translation Involving Bengali, Konkani, Maithili, Marathi, and Hindi”. 
+            Under the project titled, <b>“National Language Translation Mission (NLTM): BHASHINI”</b> 
+            funded by the Ministry of Electronics and Information Technology (MeitY), Government of India.
+        </p>
+    </div>
+    """)
+
+    # Direction dropdown
+    direction = gr.Dropdown(
+        choices=[
+            "Hindi → Konkani",
+            "Konkani → Hindi",
+        ],
+        value="Hindi → Konkani",
+        label="Translation Direction",
+    )
+
+    # Input textbox
+    input_text = gr.Textbox(
+        label="Input",
+        placeholder="Type or paste text here...",
+        lines=6,
+        max_lines=12,
+    )
+
+    # Translate button
+    translate_btn = gr.Button(
+        "Translate",
+        variant="primary",
+    )
+
+    # Output textbox
+    output_text = gr.Textbox(
+        label="Translation",
+        lines=6,
+        max_lines=12,
+    )
+
+    # Event handlers
+    translate_btn.click(
+        fn=translate,
+        inputs=[input_text, direction],
+        outputs=output_text,
+    )
+
+    input_text.submit(
+        fn=translate,
+        inputs=[input_text, direction],
+        outputs=output_text,
+    )
+
+# -----------------------------------------------------------------------------
+# Launch
+# -----------------------------------------------------------------------------
+
+if __name__ == "__main__":
+    demo.queue()
+    demo.launch(
+        server_name="0.0.0.0",
+        server_port=7860,
+    )