Update app.py

app.py

@@ -1,4 +1,4 @@
-from flask import Flask, request, jsonify, Response
+from flask import Flask, request, jsonify, Response, send_from_directory
 from transformers import pipeline, AutoTokenizer, AutoModelForCausalLM
 import torch
 import threading
@@ -25,7 +25,7 @@ SYSTEM_PROMPT_A1 = (
 MODEL_ID = "CygnisAI/Cygnis-Alpha-1.7B-v0.1-Instruct"
 
 # ─────────────────────────────────────────────
-# HTML PAGE (served at /)
+# HTML PAGE
 # ─────────────────────────────────────────────
 HTML_PAGE = """<!DOCTYPE html>
 <html lang="fr">
@@ -48,40 +48,29 @@ HTML_PAGE = """<!DOCTYPE html>
             font-weight: 900;
         }
         .glow { text-shadow: 0 0 40px rgba(165, 231, 255, 0.4); }
-
-        /* Pulse animation on logo */
         @keyframes pulse-glow {
             0%, 100% { filter: drop-shadow(0 0 20px rgba(34,211,238,0.4)); }
             50%       { filter: drop-shadow(0 0 45px rgba(34,211,238,0.9)); }
         }
         .logo-pulse { animation: pulse-glow 3s ease-in-out infinite; }
-
-        /* Fade-in on load */
         @keyframes fadeUp {
             from { opacity: 0; transform: translateY(24px); }
             to   { opacity: 1; transform: translateY(0); }
         }
-        .fade-up { animation: fadeUp 0.9s cubic-bezier(0.22,1,0.36,1) both; }
-        .delay-1 { animation-delay: 0.15s; }
-        .delay-2 { animation-delay: 0.30s; }
-
-        /* Status badge */
+        .fade-up  { animation: fadeUp 0.9s cubic-bezier(0.22,1,0.36,1) both; }
+        .delay-1  { animation-delay: 0.15s; }
+        .delay-2  { animation-delay: 0.30s; }
         @keyframes blink {
             0%, 100% { opacity: 1; }
             50%       { opacity: 0.3; }
         }
         .blink { animation: blink 1.8s ease-in-out infinite; }
-
-        /* Subtle scanline overlay */
         body::after {
             content: '';
             position: fixed; inset: 0;
             background: repeating-linear-gradient(
-                0deg,
-                transparent,
-                transparent 2px,
-                rgba(0,0,0,0.03) 2px,
-                rgba(0,0,0,0.03) 4px
+                0deg, transparent, transparent 2px,
+                rgba(0,0,0,0.03) 2px, rgba(0,0,0,0.03) 4px
             );
             pointer-events: none;
             z-index: 100;
@@ -89,15 +78,9 @@ HTML_PAGE = """<!DOCTYPE html>
     </style>
 </head>
 <body class="text-[#a5e7ff] min-h-screen flex flex-col items-center justify-center relative overflow-hidden">
-
-    <!-- Carbon texture background -->
     <div class="absolute inset-0 opacity-5"
          style="background-image:url('https://www.transparenttextures.com/patterns/carbon-fibre.png')"></div>
-
-    <!-- Main content -->
     <div class="relative z-10 flex flex-col items-center justify-center gap-10 px-6">
-
-        <!-- Logo + Title row -->
         <div class="flex flex-row items-center justify-center gap-12 fade-up">
             <div class="w-36 h-36 transition-transform duration-700 hover:scale-105">
                 <img src="/favicon.png" alt="CygnisAI Logo"
@@ -108,15 +91,11 @@ HTML_PAGE = """<!DOCTYPE html>
                 <span class="opacity-80">Console</span>
             </h1>
         </div>
-
-        <!-- Status badge -->
         <div class="fade-up delay-1 flex items-center gap-2 border border-white/10 rounded-full px-5 py-2
                     bg-white/5 backdrop-blur-sm text-sm tracking-widest uppercase text-[#a5e7ff]/70">
             <span class="blink w-2 h-2 rounded-full bg-emerald-400 inline-block"></span>
             Cygnis-Alpha · Online
         </div>
-
-        <!-- Endpoint info card -->
         <div class="fade-up delay-2 border border-white/10 rounded-2xl bg-white/5 backdrop-blur-sm
                     px-8 py-6 text-sm max-w-md w-full text-left space-y-3">
             <p class="text-white/40 uppercase tracking-widest text-xs mb-4">API Endpoint</p>
@@ -129,12 +108,11 @@ HTML_PAGE = """<!DOCTYPE html>
             </div>
         </div>
     </div>
-
 </body>
 </html>"""
 
 # ─────────────────────────────────────────────
-# MODEL — lazy-loaded in a background thread
+# MODEL — lazy-loaded in background thread
 # ─────────────────────────────────────────────
 pipe = None
 model_ready = False
@@ -143,48 +121,32 @@ model_error = None
 def load_model():
     global pipe, model_ready, model_error
     try:
-        print(f"[CygnisAI] Loading model {MODEL_ID} …")
-
-        # Use int8 quantization when bitsandbytes is available, else float32
-        try:
-            import bitsandbytes  # noqa: F401
-            print("[CygnisAI] bitsandbytes found — using load_in_8bit")
-            model = AutoModelForCausalLM.from_pretrained(
-                MODEL_ID,
-                load_in_8bit=True,
-                device_map="cpu",
-                low_cpu_mem_usage=True,
-            )
-        except ImportError:
-            print("[CygnisAI] bitsandbytes not found — using float32")
-            model = AutoModelForCausalLM.from_pretrained(
-                MODEL_ID,
-                torch_dtype=torch.float32,
-                device_map="cpu",
-                low_cpu_mem_usage=True,
-            )
+        print(f"[CygnisAI] Loading model {MODEL_ID} ...")
 
         tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
-
-        # Set pad token to avoid warnings
         if tokenizer.pad_token is None:
             tokenizer.pad_token = tokenizer.eos_token
 
+        model = AutoModelForCausalLM.from_pretrained(
+            MODEL_ID,
+            torch_dtype=torch.float32,   # float32 = stable on CPU
+            low_cpu_mem_usage=True,       # stream weights, avoid RAM spike
+        )
+        model.eval()  # disable dropout, slightly faster inference
+
         pipe = pipeline(
             "text-generation",
             model=model,
             tokenizer=tokenizer,
-            device_map="cpu",
         )
 
         model_ready = True
-        print("[CygnisAI] ✅ Model ready.")
+        print("[CygnisAI] Model ready.")
     except Exception as e:
         model_error = str(e)
-        print(f"[CygnisAI] ❌ Model load failed: {e}")
+        print(f"[CygnisAI] Model load failed: {e}")
 
 
-# Start loading in background so the server answers immediately
 threading.Thread(target=load_model, daemon=True).start()
 
 
@@ -192,15 +154,22 @@ threading.Thread(target=load_model, daemon=True).start()
 # ROUTES
 # ─────────────────────────────────────────────
 
+@app.route("/favicon.png")
+def favicon():
+    """Serve favicon.png from the project root."""
+    return send_from_directory(
+        os.path.dirname(os.path.abspath(__file__)),
+        "favicon.png",
+        mimetype="image/png"
+    )
+
 @app.route("/", methods=["GET"])
 def home():
-    """Serves the CygnisAI Console HTML page."""
     return Response(HTML_PAGE, mimetype="text/html")
 
 
 @app.route("/health", methods=["GET"])
 def health():
-    """Quick health / readiness probe."""
     if model_error:
         return jsonify({"status": "error", "detail": model_error}), 500
     if not model_ready:
@@ -210,13 +179,11 @@ def health():
 
 @app.route("/generate", methods=["POST"])
 def generate():
-    # ── Guard: model not ready yet ──────────────────────────────────────────
     if model_error:
         return jsonify({"error": f"Model failed to load: {model_error}"}), 500
     if not model_ready:
         return jsonify({"error": "Model is still loading, please retry in a moment."}), 503
 
-    # ── Parse request ────────────────────────────────────────────────────────
     data = request.get_json(silent=True)
     if not data:
         return jsonify({"error": "Request body must be valid JSON."}), 400
@@ -225,19 +192,16 @@ def generate():
     if not user_prompt:
         return jsonify({"error": "Field 'prompt' is required and cannot be empty."}), 400
 
-    # Optional params with safe defaults
-    max_new_tokens = min(int(data.get("max_new_tokens", 512)), 1024)  # cap at 1024
+    max_new_tokens = min(int(data.get("max_new_tokens", 512)), 1024)
     temperature    = float(data.get("temperature", 0.7))
     top_p          = float(data.get("top_p", 0.9))
 
-    # ── Build ChatML prompt ──────────────────────────────────────────────────
     full_prompt = (
         f"<|im_start|>system\n{SYSTEM_PROMPT_A1}<|im_end|>\n"
         f"<|im_start|>user\n{user_prompt}<|im_end|>\n"
         f"<|im_start|>assistant\n"
     )
 
-    # ── Inference ────────────────────────────────────────────────────────────
     try:
         t0 = time.time()
         output = pipe(
@@ -253,10 +217,7 @@ def generate():
         )
         elapsed = round(time.time() - t0, 2)
 
-        # Clean response: stop at im_end token
         response_text = output[0]["generated_text"].split("<|im_end|>")[0].strip()
-
-        # Fallback if model returns empty
         if not response_text:
             response_text = "Je suis Cygnis-Alpha. Comment puis-je vous aider ?"