Create app.py

app.py

@@ -0,0 +1,103 @@
+import torch
+import torch.nn.functional as F
+from transformers import AutoTokenizer, AutoModelForSequenceClassification, AutoModelForSeq2SeqLM
+from sentence_transformers import SentenceTransformer
+import gradio as gr
+import faiss
+import numpy as np
+
+# Load models
+lang_detect_model = AutoModelForSequenceClassification.from_pretrained("papluca/xlm-roberta-base-language-detection")
+lang_detect_tokenizer = AutoTokenizer.from_pretrained("papluca/xlm-roberta-base-language-detection")
+trans_model = AutoModelForSeq2SeqLM.from_pretrained("facebook/nllb-200-distilled-600M")
+trans_tokenizer = AutoTokenizer.from_pretrained("facebook/nllb-200-distilled-600M")
+embed_model = SentenceTransformer("sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2")
+
+# Language mappings
+id2lang = lang_detect_model.config.id2label
+xlm_to_nllb = {
+    "en": "eng_Latn", "fr": "fra_Latn", "hi": "hin_Deva", "es": "spa_Latn", "de": "deu_Latn",
+    "ta": "tam_Taml", "te": "tel_Telu", "ja": "jpn_Jpan", "zh": "zho_Hans", "ar": "arb_Arab",
+    "sa": "san_Deva"
+}
+nllb_langs = {
+    "eng_Latn": "English", "fra_Latn": "French", "hin_Deva": "Hindi",
+    "spa_Latn": "Spanish", "deu_Latn": "German", "tam_Taml": "Tamil",
+    "tel_Telu": "Telugu", "jpn_Jpan": "Japanese", "zho_Hans": "Chinese",
+    "arb_Arab": "Arabic", "san_Deva": "Sanskrit"
+}
+
+# Sample knowledge corpus
+corpus = [
+    "धर्म एव हतो हन्ति धर्मो रक्षति रक्षितः",
+    "Dharma when destroyed, destroys; when protected, protects.",
+    "The moon affects tides and mood, according to Jyotisha",
+    "One should eat according to the season – Rituacharya",
+    "Balance of Tridosha is health – Ayurveda principle",
+    "Ethics in Mahabharata reflect situational dharma",
+    "Meditation improves memory and mental clarity",
+    "Jyotisha links planetary motion with life patterns"
+]
+
+# Semantic index
+corpus_embeddings = embed_model.encode(corpus, convert_to_numpy=True)
+dimension = corpus_embeddings.shape[1]
+index = faiss.IndexFlatL2(dimension)
+index.add(corpus_embeddings)
+
+# Language Detection
+def detect_language(text):
+    inputs = lang_detect_tokenizer(text, return_tensors="pt", truncation=True, padding=True)
+    with torch.no_grad():
+        outputs = lang_detect_model(**inputs)
+        probs = F.softmax(outputs.logits, dim=1)
+        pred = torch.argmax(probs, dim=1).item()
+    return id2lang[pred]
+
+# Translation
+def translate(text, src_code, tgt_code):
+    trans_tokenizer.src_lang = src_code
+    encoded = trans_tokenizer(text, return_tensors="pt", truncation=True, padding=True)
+    target_lang_id = trans_tokenizer.convert_tokens_to_ids([tgt_code])[0]
+    generated = trans_model.generate(**encoded, forced_bos_token_id=target_lang_id)
+    return trans_tokenizer.decode(generated[0], skip_special_tokens=True)
+
+# Semantic Search
+def search_semantic(query, top_k=3):
+    query_embedding = embed_model.encode([query])
+    distances, indices = index.search(query_embedding, top_k)
+    results = []
+    for i, idx in enumerate(indices[0]):
+        results.append(f"{i+1}. {corpus[idx]} (Score: {distances[0][i]:.2f})")
+    return "\n".join(results)
+
+# Main function
+def pipeline(text, target_lang_code):
+    if not text.strip():
+        return "Empty input", "", "", ""
+    detected = detect_language(text)
+    src_code = xlm_to_nllb.get(detected, "eng_Latn")
+    translated = translate(text, src_code, target_lang_code)
+    matches = search_semantic(translated)
+    return text, detected, translated, matches
+
+# Language Dropdown
+lang_choices = list(nllb_langs.keys())
+
+# Gradio UI
+iface = gr.Interface(
+    fn=pipeline,
+    inputs=[
+        gr.Textbox(label="Enter your sentence"),
+        gr.Dropdown(choices=lang_choices, value="san_Deva", label="Target Language")
+    ],
+    outputs=[
+        gr.Textbox(label="Input"),
+        gr.Textbox(label="Detected Language"),
+        gr.Textbox(label="Translated Output"),
+        gr.Textbox(label="Semantic Matches")
+    ],
+    title="🌍 Sanskrit Translator + Semantic Search"
+)
+
+iface.launch()