Update app.py

app.py

@@ -67,122 +67,170 @@
 #     demo.launch()
 
 
-import gradio as gr
-from langdetect import detect
-from transformers import pipeline
-from qdrant_client import QdrantClient
-from qdrant_client.models import VectorParams, Distance
-from langchain.llms import HuggingFacePipeline
-from langchain.chains import RetrievalQA
-from langchain.vectorstores import Qdrant
-from transformers import GenerationConfig, AutoTokenizer, AutoModelForCausalLM
-from langchain.embeddings import HuggingFaceEmbeddings
-import os
+# import gradio as gr
+# from langdetect import detect
+# from transformers import pipeline
+# from qdrant_client import QdrantClient
+# from qdrant_client.models import VectorParams, Distance
+# from langchain.llms import HuggingFacePipeline
+# from langchain.chains import RetrievalQA
+# from langchain.vectorstores import Qdrant
+# from transformers import GenerationConfig, AutoTokenizer, AutoModelForCausalLM
+# from langchain.embeddings import HuggingFaceEmbeddings
+# import os
 
-QDRANT_API_KEY = os.getenv("QDRANT_API_KEY")
-QDRANT_URL = os.getenv("QDRANT_URL")
+# QDRANT_API_KEY = os.getenv("QDRANT_API_KEY")
+# QDRANT_URL = os.getenv("QDRANT_URL")
 
 
+# # Define model path
+# model_name = "FreedomIntelligence/Apollo-7B"
 
+# # Load model directly
+# tokenizer = AutoTokenizer.from_pretrained(model_name)
+# model = AutoModelForCausalLM.from_pretrained(model_name)
 
-# Define model path
-model_name = "FreedomIntelligence/Apollo-7B"
+# # Enable padding token if missing
+# tokenizer.pad_token = tokenizer.eos_token
 
-# Load model directly
-tokenizer = AutoTokenizer.from_pretrained(model_name)
-model = AutoModelForCausalLM.from_pretrained(model_name)
+# # Set up Qdrant vector store
+# qdrant_client = QdrantClient(url=QDRANT_URL, api_key = QDRANT_API_KEY)
+# vector_size = 768
+# embedding = HuggingFaceEmbeddings(model_name="Omartificial-Intelligence-Space/GATE-AraBert-v1")
 
-# Enable padding token if missing
-tokenizer.pad_token = tokenizer.eos_token
+# qdrant_vectorstore = Qdrant(
+#     client=qdrant_client,
+#     collection_name="arabic_rag_collection",
+#     embeddings=embedding
+# )
 
-# Set up Qdrant vector store
-qdrant_client = QdrantClient(url=QDRANT_URL, api_key = QDRANT_API_KEY)
-vector_size = 768
-embedding = HuggingFaceEmbeddings(model_name="Omartificial-Intelligence-Space/GATE-AraBert-v1")
+# # Generation config
+# generation_config = GenerationConfig(
+#     max_new_tokens=150,
+#     temperature=0.2,
+#     top_k=20,
+#     do_sample=True,
+#     top_p=0.7,
+#     repetition_penalty=1.3,
+# )
 
-qdrant_vectorstore = Qdrant(
-    client=qdrant_client,
-    collection_name="arabic_rag_collection",
-    embeddings=embedding
-)
+# # Set up HuggingFace Pipeline
+# llm_pipeline = pipeline(
+#     model=model,
+#     tokenizer=tokenizer,
+#     task="text-generation",
+#     generation_config=generation_config,
+# )
 
-# Generation config
-generation_config = GenerationConfig(
-    max_new_tokens=150,
-    temperature=0.2,
-    top_k=20,
-    do_sample=True,
-    top_p=0.7,
-    repetition_penalty=1.3,
-)
+# llm = HuggingFacePipeline(pipeline=llm_pipeline)
 
-# Set up HuggingFace Pipeline
-llm_pipeline = pipeline(
-    model=model,
-    tokenizer=tokenizer,
-    task="text-generation",
-    generation_config=generation_config,
+# # Set up QA Chain
+# qa_chain = RetrievalQA.from_chain_type(
+#     llm=llm,
+#     retriever=qdrant_vectorstore.as_retriever(search_kwargs={"k": 3}),
+#     chain_type="stuff"
+# )
+
+# # Generate prompt based on language
+# def generate_prompt(question):
+#     lang = detect(question)
+#     if lang == "ar":
+#         return f"""أجب على السؤال الطبي التالي بلغة عربية فصحى، بإجابة دقيقة ومفصلة. إذا لم تجد معلومات كافية في السياق، استخدم معرفتك الطبية السابقة. 
+#  وتأكد من ان:
+# - عدم تكرار أي نقطة أو عبارة أو كلمة
+# - وضوح وسلاسة كل نقطة
+# - تجنب الحشو والعبارات الزائدة-
+
+# السؤال: {question}
+# الإجابة:
+# """
+#     else:
+#         return f"""Answer the following medical question in clear English with a detailed, non-redundant response. Do not repeat ideas, phrases, or restate the question in the answer. If the context lacks relevant information, rely on your prior medical knowledge. If the answer involves multiple points, list them in concise and distinct bullet points:
+# Question: {question}
+# Answer:"""
+
+# # Define Gradio interface function
+# def medical_chatbot(question):
+#     formatted_question = generate_prompt(question)
+#     answer = qa_chain.run(formatted_question)
+#     return answer
+
+# # Set up Gradio interface
+# iface = gr.Interface(
+#     fn=medical_chatbot,
+#     inputs=gr.Textbox(label="Ask a Medical Question", placeholder="Type your question here..."),
+#     outputs=gr.Textbox(label="Answer", interactive=False),
+#     title="Medical Chatbot",
+#     description="Ask medical questions and get detailed answers in Arabic or English.",
+#     theme="compact"
+# )
+
+# # Launch Gradio interface
+# if __name__ == "__main__":
+#     iface.launch()
+
+
+import gradio as gr
+from langdetect import detect
+from transformers import AutoTokenizer, AutoModelForCausalLM, TextGenerationPipeline
+import torch
+
+# Load model and tokenizer
+model_name = "FreedomIntelligence/Apollo-7B"
+
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+model = AutoModelForCausalLM.from_pretrained(
+    model_name,
+    torch_dtype=torch.float16,
+    device_map="auto"
 )
 
-llm = HuggingFacePipeline(pipeline=llm_pipeline)
+tokenizer.pad_token = tokenizer.eos_token
 
-# Set up QA Chain
-qa_chain = RetrievalQA.from_chain_type(
-    llm=llm,
-    retriever=qdrant_vectorstore.as_retriever(search_kwargs={"k": 3}),
-    chain_type="stuff"
+# Create generation pipeline
+pipe = TextGenerationPipeline(
+    model=model,
+    tokenizer=tokenizer,
+    device=model.device.index if torch.cuda.is_available() else -1
 )
 
-# Generate prompt based on language
-def generate_prompt(question):
-    lang = detect(question)
+# Prompt formatter based on language
+def generate_prompt(message, history):
+    lang = detect(message)
     if lang == "ar":
-        return f"""أجب على السؤال الطبي التالي بلغة عربية فصحى، بإجابة دقيقة ومفصلة. إذا لم تجد معلومات كافية في السياق، استخدم معرفتك الطبية السابقة. 
+        return f"""أجب على السؤال الطبي التالي بلغة عربية فصحى، بإجابة دقيقة ومفصلة. إذا لم تجد معلومات كافية في السياق، استخدم معرفتك الطبية السابقة.
  وتأكد من ان:
 - عدم تكرار أي نقطة أو عبارة أو كلمة
 - وضوح وسلاسة كل نقطة
-- تجنب الحشو والعبارات الزائدة-
+- تجنب الحشو والعبارات الزائدة
 
-السؤال: {question}
-الإجابة:
-"""
+السؤال: {message}
+الإجابة:"""
     else:
-        return f"""Answer the following medical question in clear English with a detailed, non-redundant response. Do not repeat ideas, phrases, or restate the question in the answer. If the context lacks relevant information, rely on your prior medical knowledge. If the answer involves multiple points, list them in concise and distinct bullet points:
-Question: {question}
+        return f"""Answer the following medical question in clear English with a detailed, non-redundant response. Do not repeat ideas or restate the question. If information is missing, rely on your prior medical knowledge:
+Question: {message}
 Answer:"""
 
-# Define Gradio interface function
-def medical_chatbot(question):
-    formatted_question = generate_prompt(question)
-    answer = qa_chain.run(formatted_question)
+# Chat function
+def chat_fn(message, history):
+    prompt = generate_prompt(message, history)
+    response = pipe(prompt, max_new_tokens=512, temperature=0.7, top_p=0.9)[0]['generated_text']
+    answer = response.split("Answer:")[-1].strip() if "Answer:" in response else response.split("الإجابة:")[-1].strip()
     return answer
 
-# Set up Gradio interface
-iface = gr.Interface(
-    fn=medical_chatbot,
-    inputs=gr.Textbox(label="Ask a Medical Question", placeholder="Type your question here..."),
-    outputs=gr.Textbox(label="Answer", interactive=False),
-    title="Medical Chatbot",
-    description="Ask medical questions and get detailed answers in Arabic or English.",
+# Gradio ChatInterface
+demo = gr.ChatInterface(
+    fn=chat_fn,
+    title="🩺 Apollo-7B Medical Chatbot",
+    description="Multilingual (Arabic & English) medical Q&A chatbot powered by Apollo-7B. No RAG, just fast model inference.",
+    examples=[
+        "ما هي أعراض ضغط الدم المرتفع؟",
+        "What are the side effects of paracetamol?",
+        "هل يمكن علاج مرض السكري؟",
+        "How does COVID-19 affect the lungs?"
+    ],
     theme="compact"
 )
 
-# demo = gr.ChatInterface(
-#     respond,
-#     additional_inputs=[
-#         gr.Textbox(value="You are a Medical Chatbot.", label="System message"),
-#         gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
-#         gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
-#         gr.Slider(
-#             minimum=0.1,
-#             maximum=1.0,
-#             value=0.95,
-#             step=0.05,
-#             label="Top-p (nucleus sampling)",
-#         ),
-#     ],
-# )
-
-# Launch Gradio interface
 if __name__ == "__main__":
-    iface.launch()
+    demo.launch()