Update app.py

app.py

@@ -1,65 +1,67 @@
 import gradio as gr
-from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
-import json
+import requests
+import os
+import requests
 
-# Load words and languages from JSON files
+# Gradio-compatible data loading (as before)
 with open("top_500_quran_lemmas_fixed.json", encoding="utf-8") as f:
     word_list = json.load(f)
 
 with open("language_list.json", encoding="utf-8") as f:
     language_list = json.load(f)
 
-# Format dropdown options
 word_options = [f"{word['text']} ({word['english']})" for word in word_list]
 language_options = [f"{lang['name']} ({lang['code']})" for lang in language_list]
 
-# Load DeepSeek-V3 model
-model_id = "deepseek-ai/DeepSeek-V3"
-tokenizer = AutoTokenizer.from_pretrained(model_id)
-model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype="auto", device_map="auto")
-generator = pipeline("text-generation", model=model, tokenizer=tokenizer)
+# Use Hugging Face Inference API
+API_URL = "https://api-inference.huggingface.co/models/deepseek-ai/DeepSeek-V3"
+HEADERS = {"Authorization": f"Bearer {os.getenv('HF_API_TOKEN')}"}
+
+def query_hf_model(payload):
+    response = requests.post(API_URL, headers=HEADERS, json=payload)
+    return response.json()
 
-# Generate prompt
 def create_prompt(word_entry, language_code):
-    prompt = f"""
+    return f"""
 You are a friendly Quranic AI assistant.
 
-The word is: {word_entry['text']} ({word_entry['english']})
+The word is: {word_entry['text']} ({word_entry['english']}).
 
-Please provide the following in simple, easy-to-understand language, translated into {language_code}:
+Please provide the following in easy-to-understand {language_code}:
 1. Translation of the word.
-2. Its root word and any related words (derivatives).
-3. Where it appears in the Qur'an — list the Surah and Ayah numbers.
-4. Give an explanation of each appearance (based on context), in {language_code}.
+2. Its root word and any related derivatives.
+3. Where it appears in the Qur'an (surah and ayah).
+4. Simple explanation for each occurrence.
 
-Avoid technical terms. Make it feel like a helpful teacher explaining to a student.
-"""
-    return prompt.strip()
+Avoid technical terms. Think like a teacher explaining clearly to students.
+""".strip()
 
-# Function to call the model
 def process(word_label, lang_label):
-    # Extract selected word data
     selected_word = next((w for w in word_list if w['text'] in word_label), None)
     language_code = lang_label.split("(")[-1].strip(")")
-
     if not selected_word:
-        return "Word not found in list."
+        return "Word not found."
 
     prompt = create_prompt(selected_word, language_code)
-    result = generator(prompt, max_new_tokens=800, do_sample=True, temperature=0.7)[0]["generated_text"]
+    output = query_hf_model({"inputs": prompt})
 
-    return result.replace(prompt, "").strip()
+    # Some formatting safety
+    if isinstance(output, dict) and "error" in output:
+        return f"Error: {output['error']}"
+    elif isinstance(output, list):
+        return output[0].get("generated_text", "No response.")
+    else:
+        return str(output)
 
 # Gradio UI
 with gr.Blocks() as demo:
-    gr.Markdown("## 📖 Quran Word Explorer with DeepSeek-V3")
+    gr.Markdown("## 📖 Quran Word Explorer using DeepSeek-V3 (HF Inference API)")
     with gr.Row():
         word_input = gr.Dropdown(choices=word_options, label="Select a Quran Word")
         lang_input = gr.Dropdown(choices=language_options, label="Select Language")
 
     output = gr.Textbox(label="DeepSeek-V3 Output", lines=20)
     run_btn = gr.Button("Get Info")
-
     run_btn.click(fn=process, inputs=[word_input, lang_input], outputs=output)
 
 demo.launch()