Update app.py

app.py

@@ -12,7 +12,7 @@ try:
 except Exception:
     ZoneInfo = None  # graceful fallback to UTC
 
-# Try to import Cohere SDK if present (for hosted path)
+# Try Cohere SDK if present (for hosted path)
 try:
     import cohere  # pip install cohere
     _HAS_COHERE = True
@@ -22,7 +22,6 @@ except Exception:
 from transformers import AutoTokenizer, AutoModelForCausalLM
 from huggingface_hub import login, HfApi
 
-
 # -------------------
 # Configuration
 # -------------------
@@ -36,9 +35,8 @@ HF_TOKEN = (
 COHERE_API_KEY = os.getenv("COHERE_API_KEY")
 USE_HOSTED_COHERE = bool(COHERE_API_KEY and _HAS_COHERE)
 
-
 # -------------------
-# Helpers (used for connection/status only)
+# Helpers (status only)
 # -------------------
 def local_now_str(user_tz: str | None) -> tuple[str, str]:
     """Returns (label, formatted_time). Falls back to UTC if tz missing/invalid."""
@@ -62,7 +60,6 @@ def pick_dtype_and_map():
         return torch.float16, {"": "mps"}
     return torch.float32, "cpu"  # CPU path (likely too big for R7B)
 
-
 def is_identity_query(message: str, history) -> bool:
     """Detects identity questions in current message or most recent user turn."""
     patterns = [
@@ -77,23 +74,17 @@ def is_identity_query(message: str, history) -> bool:
         r"\byour\s+name\b",
         r"\bwho\s+am\s+i\s+chatting\s+with\b",
     ]
-
     def hit(text: str | None) -> bool:
         t = (text or "").strip().lower()
         return any(re.search(p, t) for p in patterns)
-
     if hit(message):
         return True
-
     if history:
-        # Gradio history: List[Tuple[user, assistant]]
         last_user = history[-1][0] if isinstance(history[-1], (list, tuple)) and history[-1] else None
         if hit(last_user):
             return True
-
     return False
 
-
 # -------------------
 # Cohere Hosted Path
 # -------------------
@@ -101,7 +92,6 @@ _co_client = None
 if USE_HOSTED_COHERE:
     _co_client = cohere.Client(api_key=COHERE_API_KEY)
 
-
 def _cohere_parse(resp):
     # v5+ responses.create
     if hasattr(resp, "output_text") and resp.output_text:
@@ -115,7 +105,6 @@ def _cohere_parse(resp):
         return resp.text.strip()
     return "Sorry, I couldn't parse the response from Cohere."
 
-
 def cohere_chat(message, history):
     try:
         # Prefer modern API
@@ -143,7 +132,6 @@ def cohere_chat(message, history):
     except Exception as e:
         return f"Error calling Cohere API: {e}"
 
-
 # -------------------
 # Local HF Path
 # -------------------
@@ -154,9 +142,7 @@ def load_local_model():
             "HUGGINGFACE_HUB_TOKEN (or HF_TOKEN) is not set. "
             "Either set it, or provide COHERE_API_KEY to use Cohere's hosted API."
         )
-
     login(token=HF_TOKEN, add_to_git_credential=False)
-
     dtype, device_map = pick_dtype_and_map()
     tok = AutoTokenizer.from_pretrained(
         MODEL_ID,
@@ -178,7 +164,6 @@ def load_local_model():
         mdl.config.eos_token_id = tok.eos_token_id
     return mdl, tok
 
-
 def build_inputs(tokenizer, message, history):
     msgs = []
     for u, a in (history or []):
@@ -189,7 +174,6 @@ def build_inputs(tokenizer, message, history):
         msgs, tokenize=True, add_generation_prompt=True, return_tensors="pt"
     )
 
-
 def local_generate(model, tokenizer, input_ids, max_new_tokens=350):
     input_ids = input_ids.to(model.device)
     with torch.no_grad():
@@ -207,23 +191,18 @@ def local_generate(model, tokenizer, input_ids, max_new_tokens=350):
     text = tokenizer.decode(gen_only, skip_special_tokens=True)
     return text.strip()
 
-
 # -------------------
-# Chat callback (no header/meta in chat replies)
+# Chat callback (no meta in replies)
 # -------------------
 def chat_fn(message, history, user_tz):
     try:
-        # Identity override → return ONLY the brand line
         if is_identity_query(message, history):
             return "I am ClarityOps, your strategic decision making AI partner."
-
         if USE_HOSTED_COHERE:
             return cohere_chat(message, history)
-
         model, tokenizer = load_local_model()
         inputs = build_inputs(tokenizer, message, history)
         return local_generate(model, tokenizer, inputs, max_new_tokens=350)
-
     except RuntimeError as e:
         emsg = str(e)
         if "out of memory" in emsg.lower() or "cuda" in emsg.lower():
@@ -232,9 +211,8 @@ def chat_fn(message, history, user_tz):
     except Exception as e:
         return f"Error during chat: {e}"
 
-
 # -------------------
-# THEME & STYLES (compatible with your Gradio)
+# Theme & Styles (compatible with broad Gradio versions)
 # -------------------
 theme = gr.themes.Soft(
     primary_hue="teal",
@@ -247,13 +225,12 @@ theme = gr.themes.Soft(
 
 custom_css = """
 :root {
-  --brand-bg: #f6fbfb;
+  --brand-bg: #e6f7f8;        /* soft medical teal */
   --brand-card: #ffffff;
-  --brand-text: #0f172a; /* slate-900 */
-  --brand-subtle: #475569; /* slate-600 */
-  --brand-accent: #0d9488; /* teal-600 */
-  --brand-accent-weak: #99f6e4; /* teal-200 */
-  --brand-border: #e2e8f0; /* slate-200 */
+  --brand-text: #0f172a;      /* slate-900 */
+  --brand-subtle: #475569;    /* slate-600 */
+  --brand-accent: #0d9488;    /* teal-600 */
+  --brand-border: #cbd5e1;    /* slate-300 */
 }
 
 /* Page background */
@@ -271,9 +248,14 @@ h1, .prose h1 {
   font-size: 28px !important; /* set via CSS for compatibility */
 }
 
-/* Chat body text */
-.message {
-  font-size: 16px !important;
+/* Chat bubbles */
+.message.user {
+  background: var(--brand-accent) !important; /* teal bubble */
+  color: #ffffff !important;                  /* white text */
+}
+.message.bot {
+  background: var(--brand-card) !important;   /* white bubble */
+  color: var(--brand-text) !important;        /* dark text */
 }
 
 /* Status badge wrapper */
@@ -298,7 +280,7 @@ h1, .prose h1 {
   font-size: 14px;
 }
 
-/* Description / helper text */
+/* Helper text */
 .helper {
   color: var(--brand-subtle);
   margin: .25rem 0 1rem 0;
@@ -309,78 +291,66 @@ h1, .prose h1 {
   border-radius: 16px !important;
 }
 
-/* Chat bubbles */
-.message.user {
-  background: #f8fafc !important;
-}
-.message.bot {
-  background: #ffffff !important;
-}
-
 /* Inputs */
 textarea, input, .gr-input {
   border-radius: 12px !important;
 }
 """
 
-
 # -------------------
 # UI
 # -------------------
 with gr.Blocks(theme=theme, css=custom_css) as demo:
-    # Hidden textbox to hold browser timezone (Gradio expects components for outputs)
+    # Hidden textbox to hold browser timezone
     tz_box = gr.Textbox(visible=False)
 
-    # On load, capture browser timezone via JS and write it into tz_box
+    # Capture browser timezone via JS and store in tz_box
     demo.load(
-        fn=lambda tz: tz,                     # echo JS value to Python
-        inputs=[tz_box],                      # 1 input required for lambda
-        outputs=[tz_box],                     # write into same hidden box
+        fn=lambda tz: tz,   # echo JS value
+        inputs=[tz_box],
+        outputs=[tz_box],
         js="() => Intl.DateTimeFormat().resolvedOptions().timeZone"
     )
 
-    # Model status (auto, no button)
+    # Model status (auto, one-line badge)
     def model_status(_user_tz):
         try:
             if USE_HOSTED_COHERE:
                 return (
                     '<div class="status-wrap">'
-                    '<span class="badge">✅ Connected • Cohere API — model: <strong>command-r7b-12-2024</strong></span>'
-                    "</div>"
+                    '<span class="badge">✅ Connected • Cohere API — model: '
+                    '<strong>command-r7b-12-2024</strong></span></div>'
                 )
             api = HfApi(token=HF_TOKEN)
             mi = api.model_info(MODEL_ID)
             return (
                 '<div class="status-wrap">'
-                f'<span class="badge">✅ Connected • Local HF — model: <strong>{mi.modelId}</strong></span>'
-                "</div>"
+                f'<span class="badge">✅ Connected • Local HF — model: '
+                f'<strong>{mi.modelId}</strong></span></div>'
             )
         except Exception as e:
             return (
                 '<div class="status-wrap">'
                 f'<span class="badge" style="background:#fff7ed;color:#9a3412;border-color:#fed7aa;">'
-                f'⚠️ Connection Issue — {str(e)}'
-                '</span></div>'
+                f'⚠️ Connection Issue — {str(e)}</span></div>'
             )
 
-    # Header
+    # Header + status
     gr.Markdown("# Medical Decision Support AI")
-
-    # Status line (renders HTML badge)
     status_line = gr.HTML("<div class='status-wrap'><span class='badge'>Connecting…</span></div>")
     demo.load(fn=model_status, inputs=[tz_box], outputs=[status_line])
 
-    # Subtle helper text
+    # Helper text
     gr.Markdown(
         "<div class='helper'>Designed for healthcare executives: concise, reliable decision support. "
         "First response may take a moment while the model warms up.</div>"
     )
 
     # Chat
-    chat = gr.ChatInterface(
+    gr.ChatInterface(
         fn=chat_fn,
         type="messages",
-        additional_inputs=[tz_box],  # pass timezone into chat_fn
+        additional_inputs=[tz_box],  # pass timezone into chat_fn (future use)
         description="",
         examples=[
             ["What are the symptoms of hypertension?", ""],
@@ -395,3 +365,4 @@ if __name__ == "__main__":
 
 
 
+