Update app.py

app.py

@@ -1,70 +1,62 @@
 import os
 import streamlit as st
-from huggingface_hub import InferenceApi
+import requests
 
 # ── Config ────────────────────────────────────────────────────────────────
-MODEL_ID       = "sourize/phi2-memory-lora"
-HF_TOKEN       = os.getenv("HF_TOKEN")
-CONTEXT_TURNS  = 6
-MAX_NEW_TOKENS = 128
-TEMPERATURE    = 0.7
-TOP_P          = 0.9
+HF_TOKEN      = os.getenv("HF_TOKEN") 
+MODEL_ID      = "sourize/phi2-memory-lora"
+CONTEXT_TURNS = 6
 
 SYSTEM_PROMPT = (
     "You are a helpful assistant for DeepTalks with base Phi-2\n"
     "fine-tuned by Sourish for domain support.\n"
     "Answer **only** using the conversation context below.\n"
-    "Do NOT output lines beginning with 'User:' or 'Assistant:'.\n"
+    "Do NOT output any lines beginning with 'User:' or 'Assistant:'.\n"
     "If you don't know, say \"I don't know.\"\n"
 )
 
-# ── Inference API ─────────────────────────────────────────────────────────
-inference = InferenceApi(
-    repo_id=MODEL_ID,
-    token=HF_TOKEN,
-    task="text-generation"
-)
+API_URL = f"https://api-inference.huggingface.co/models/{MODEL_ID}"
+HEADERS = {"Authorization": f"Bearer {HF_TOKEN}"}
 
 def query_hf(prompt: str) -> str:
-    # All generation kwargs go at top level
-    outputs = inference(
-        inputs=prompt,
-        max_new_tokens=MAX_NEW_TOKENS,
-        do_sample=True,
-        temperature=TEMPERATURE,
-        top_p=TOP_P,
-        return_full_text=False,
-    )
-    # returns a list of dicts: [{"generated_text": "..."}]
-    text = outputs[0]["generated_text"].strip()
-    # strip stray speaker markers
-    for marker in ("User:", "Assistant:"):
-        if marker in text:
-            text = text.split(marker)[0].strip()
-    return text or "I don't know."
+    payload = {
+        "inputs": prompt,
+        "parameters": {
+            "max_new_tokens": 128,
+            "do_sample": True,
+            "temperature": 0.7,
+            "top_p": 0.9,
+            "return_full_text": False
+        },
+        "options": {"use_cache": False}
+    }
+    r = requests.post(API_URL, headers=HEADERS, json=payload, timeout=60)
+    r.raise_for_status()
+    out = r.json()[0]["generated_text"].strip()
+    for m in ("User:", "Assistant:"):
+        if m in out:
+            out = out.split(m)[0].strip()
+    return out or "I don't know."
 
 # ── Streamlit UI ──────────────────────────────────────────────────────────
 st.set_page_config(layout="centered")
-st.title("🧠 DeepTalks (Inference API)")
+st.title("🧠 DeepTalks (API)")
 st.subheader("Your personal AI Companion")
 
 if "history" not in st.session_state:
     st.session_state.history = []
 
-# render past chat
-for role, msg in st.session_state.history:
-    st.chat_message("user" if role=="You" else "assistant").write(msg)
+for role, txt in st.session_state.history:
+    st.chat_message("user" if role=="You" else "assistant").write(txt)
 
-# user input
 user_input = st.chat_input("Type your message…")
 if user_input:
     st.chat_message("user").write(user_input)
     st.session_state.history.append(("You", user_input))
 
-    # build context
     recent = st.session_state.history[-CONTEXT_TURNS*2:]
-    context = "\n".join(text for _, text in recent)
-    prompt  = (
+    context = "\n".join(msg for _, msg in recent)
+    prompt = (
         f"{SYSTEM_PROMPT}\n\n"
         f"Context:\n{context}\n\n"
         f"User: {user_input}\nAssistant:"
@@ -74,7 +66,7 @@ if user_input:
         try:
             reply = query_hf(prompt)
         except Exception as e:
-            st.error(f"Inference error: {e}")
+            st.error(f"API error: {e}")
             reply = "I’m sorry, something went wrong."
 
     st.chat_message("assistant").write(reply)