jean-router: contexte reel depuis llama-server + overrides

- Interroge llama-server:8080 pour le n_ctx du modele actif
- JEAN_CTX_OVERRIDES prioritaire si defini
- Fallback: Ollama matching → heuristique
- Plus de devinette, le contexte est precis

jean-router/app.py

@@ -38,6 +38,7 @@ ENABLE_WOL = os.getenv("ENABLE_WOL", "false").lower() == "true"
 TARGET_MAC = os.getenv("TARGET_MAC", os.getenv("PC_GAMING_MAC", ""))
 BROADCAST_IP = os.getenv("BROADCAST_IP", "192.168.1.255")
 WAKE_TIMEOUT = int(os.getenv("WAKE_TIMEOUT", "180"))
+JEAN_CTX_OVERRIDES = json_mod.loads(os.getenv("JEAN_CTX_OVERRIDES", "{}"))
 wol_in_progress = threading.Lock()
 jean_alive_cache = {"alive": False, "timestamp": 0}
 
@@ -469,9 +470,29 @@ def _match_preset_to_ollama(preset_id: str):
 
 def get_jean_metadata():
     """Resout dynamiquement context_length et capabilities pour les presets Jean.
-    1) Ollama matching → 2) Jean /v1/models → 3) heuristique"""
-    now = time.time()
+    1) JEAN_CTX_OVERRIDES → 2) llama-server /v1/models → 3) Ollama matching → 4) heuristique"""
 
+    # 1 — Overrides utilisateur (priorite max)
+    for pid, ctx in JEAN_CTX_OVERRIDES.items():
+        jean_ctx_cache[pid] = ctx
+
+    # 2 — llama-server direct (port 8080, modele actif)
+    try:
+        r = requests.get(f"http://{JEAN_WEB_HOST}:8080/v1/models", timeout=3)
+        if r.ok:
+            for m in r.json().get("data", []):
+                meta = m.get("meta", {})
+                ctx = meta.get("n_ctx") or m.get("max_model_len") or m.get("context_length")
+                if ctx:
+                    # Associer ce ctx au preset actif (meme nom ou alias)
+                    active = get_active_preset()
+                    if active:
+                        jean_ctx_cache[active["id"]] = int(ctx)
+                        log.info(f"[Jean] ctx reel via llama-server: {active['id']}={ctx}")
+    except Exception:
+        pass
+
+    # 3 — Jean proxy /v1/models (via jean-proxy, pour le modele actif)
     try:
         url = f"http://{JEAN_PROXY_HOST}:{JEAN_PROXY_PORT}/v1/models"
         r = requests.get(url, timeout=2)
@@ -480,7 +501,7 @@ def get_jean_metadata():
                 alias = m.get("id", "")
                 meta = m.get("meta", {})
                 n_ctx = meta.get("n_ctx") or m.get("max_model_len")
-                if n_ctx and alias:
+                if n_ctx and alias and alias not in jean_ctx_cache:
                     jean_ctx_cache[alias] = int(n_ctx)
     except Exception:
         pass
@@ -503,7 +524,9 @@ def get_jean_metadata():
         ollama_name = matched_ollama.get("name", "").replace(":latest", "") if matched_ollama else None
 
         if pid not in jean_ctx_cache:
-            if ollama_name and ollama_lookup.get(ollama_name, {}).get("context_length"):
+            if pid in JEAN_CTX_OVERRIDES:
+                jean_ctx_cache[pid] = JEAN_CTX_OVERRIDES[pid]
+            elif ollama_name and ollama_lookup.get(ollama_name, {}).get("context_length"):
                 jean_ctx_cache[pid] = ollama_lookup[ollama_name]["context_length"]
             else:
                 jean_ctx_cache[pid] = _guess_context_length(pid)
@@ -1202,9 +1225,8 @@ h1{font-size:18px;color:var(--purple);margin-bottom:12px}
   </div>
   <div class="card">
     <h2>⚡ Actions</h2>
-    <button class="btn" onclick="action('refresh')">🔄 Refresh cache</button>
+    <button class="btn" onclick="action('refresh');setTimeout(fetchStatus,800)">🔄 Actualiser modèles Jean + Ollama</button>
     <button class="btn btn-danger" onclick="action('restart-jean')" title="Stop + Start Jean">🔁 Restart Jean</button>
-    <button class="btn" onclick="location.reload()">↻ Reload page</button>
     <div style="margin-top:8px">
       <span class="badge badge-blue" id="wol-badge">WoL: -</span>
       <span class="badge badge-green" id="cache-badge">Cache: -</span>