Update app.py

app.py

@@ -24,6 +24,7 @@ from langchain_openai import OpenAIEmbeddings, ChatOpenAI
 from langchain_community.vectorstores import Qdrant
 from langchain_community.embeddings import HuggingFaceEmbeddings
 from langchain_community.embeddings import SentenceTransformerEmbeddings
+from transformers import pipeline, AutoTokenizer, AutoModelForSequenceClassification, AutoModelForCausalLM
 import nest_asyncio
 
 torch.cuda.empty_cache()
@@ -35,7 +36,7 @@ st.set_page_config(page_title="DermBOT", page_icon="🧬", layout="centered")
 
 
 # === Model Selection ===
-available_models = ["OpenAI GPT-4o", "LLaMA 3", "Gemini Pro"]
+available_models = ["GPT-4o", "LLaMA 4 Maverick", "Gemini 2.5 Pro","All"]
 st.session_state["selected_model"] = st.sidebar.selectbox("Select LLM Model", available_models)
 
 
@@ -130,6 +131,37 @@ elif "Gemini" in selected_model:
         return response.text
 
     llm = get_gemini_response
+    
+elif "All" in selected_model:
+
+    from groq import Groq
+    import google.generativeai as genai
+    genai.configure(api_key=st.secrets["GEMINI_API_KEY"])
+    pair_ranker = pipeline("text-classification", model="llm-blender/PairRM")
+    gen_fuser = pipeline("text-generation", model="llm-blender/gen_fuser_3b", max_length=2048, do_sample=False)
+
+    def get_all_model_responses(prompt):
+        openai_resp = ChatOpenAI(model="gpt-4o", temperature=0.2, api_key=st.secrets["OPENAI_API_KEY"]).invoke(
+            [{"role": "system", "content": prompt}]).content
+
+        gemini = genai.GenerativeModel("gemini-2.5-pro-exp-03-25")
+        gemini_resp = gemini.generate_content(prompt).text
+
+        llama = Groq(api_key=st.secrets["GROQ_API_KEY"])
+        llama_resp = llama.chat.completions.create(
+            model="meta-llama/llama-4-maverick-17b-128e-instruct",
+            messages=[{"role": "user", "content": prompt}],
+            temperature=1, max_completion_tokens=1024, top_p=1, stream=False
+        ).choices[0].message.content
+
+        return [openai_resp, gemini_resp, llama_resp]
+
+    def rank_and_fuse(prompt, responses):
+        ranked = [(resp, pair_ranker(f"{prompt}\n\n{resp}")[0][1]['score']) for resp in responses]
+        ranked.sort(key=lambda x: x[1], reverse=True)
+        fusion_input = "\n\n".join([f"[Answer {i+1}]: {ans}" for i, (ans, _) in enumerate(ranked)])
+        return gen_fuser(f"Fuse these responses:\n{fusion_input}", return_full_text=False)[0]['generated_text']
+
 
 else:
     st.error("Unsupported model selected.")
@@ -293,30 +325,50 @@ def export_chat_to_pdf(messages):
 
 
 #Reranker utility
-def rerank_with_cohere(query: str, documents: list, top_n: int = 5) -> list:
+def rerank_with_cohere(query, documents, top_n=5):
     if not documents:
         return []
-
-    raw_texts = [doc.page_content if hasattr(doc, "page_content") else str(doc) for doc in documents]
+    raw_texts = [doc.page_content for doc in documents]
     results = co.rerank(query=query, documents=raw_texts, top_n=min(top_n, len(raw_texts)), model="rerank-v3.5")
-    reranked_docs = [documents[result.index] for result in results]
-    return reranked_docs
+    return [documents[result.index] for result in results]
+
+
+
+
+
+pair_ranker = pipeline(
+    "text-classification",
+    model="llm-blender/PairRM",
+    tokenizer="llm-blender/PairRM",
+    return_all_scores=True
+)
+
+gen_fuser = pipeline(
+    "text-generation",
+    model="llm-blender/gen_fuser_3b",
+    tokenizer="llm-blender/gen_fuser_3b",
+    max_length=2048,
+    do_sample=False
+)
+
 
 # Final answer generation using reranked context
-def get_reranked_response(query: str):
+def get_reranked_response(query):
     docs = retriever.get_relevant_documents(query)
     reranked_docs = rerank_with_cohere(query, docs)
     context = "\n\n".join([doc.page_content for doc in reranked_docs])
     prompt = AI_PROMPT_TEMPLATE.format(question=query, context=context)
 
+    if selected_model == "All":
+        responses = get_all_model_responses(prompt)
+        fused = rank_and_fuse(prompt, responses)
+        return type("Obj", (), {"content": fused})
+
     if callable(llm):
-        # Gemini or LLaMA
         return type("Obj", (), {"content": llm(prompt)})
     else:
-        # OpenAI LangChain interface
         return llm.invoke([{"role": "system", "content": prompt}])
 
-
 # === App UI ===
 
 st.title("🧬 DermBOT — Skin AI Assistant")