Update app.py

app.py

@@ -9,7 +9,8 @@ from typing import Optional
 
 MODEL_REPO = os.environ.get("MODEL_REPO", "DavidAU/Qwen3.6-27B-Heretic-Uncensored-FINETUNE-NEO-CODE-Di-IMatrix-MAX-GGUF")
 MODEL_FILE = os.environ.get("MODEL_FILE", "Qwen3.6-27B-NEO-CODE-HERE-2T-OT-Q4_K_M.gguf")
-N_CTX = int(os.environ.get("N_CTX", "4096"))
+N_CTX = int(os.environ.get("N_CTX", "16384"))          # FIX: was 4096
+MAX_TOKENS = int(os.environ.get("MAX_TOKENS", "4096"))  # FIX: env-tunable
 N_GPU = int(os.environ.get("N_GPU_LAYERS", os.environ.get("DAVIDAU_N_GPU_LAYERS", "-1")))
 
 SYSTEM_PROMPT = """You are codeMax, an uncensored 27B coding assistant. You excel at:
@@ -19,8 +20,11 @@ SYSTEM_PROMPT = """You are codeMax, an uncensored 27B coding assistant. You exce
 - Reading shared files and answering about them
 Rules: put code in ```language blocks. Be precise and thorough."""
 
+CSS = """footer { display: none !important; }"""  # FIX: css lives here now
+
 _llm = None
 _model_path: Optional[str] = None
+_boot_error: Optional[str] = None   # FIX: surface startup failures in chat instead of a dead space
 
 def _download():
     global _model_path
@@ -50,8 +54,13 @@ def _load():
     print("[MODEL] Ready.", file=sys.stderr)
     return _llm
 
-print("[BOOT] Loading model at startup...", file=sys.stderr)
-_load()
+# FIX: don't let a startup OOM kill the whole space silently — serve the UI, report the error
+try:
+    print("[BOOT] Loading model at startup...", file=sys.stderr)
+    _load()
+except Exception as e:
+    _boot_error = f"{type(e).__name__}: {e}"
+    print(f"[BOOT] Model load failed (will retry on first message): {_boot_error}", file=sys.stderr)
 
 def _read_text(path, enc="utf-8"):
     for e in [enc, "utf-8-sig", "latin-1", "cp1252", "utf-16"]:
@@ -158,6 +167,17 @@ def parse_all_files(files):
     return "\n".join(parts)
 
 def respond(message, history, uploaded_files):
+    # FIX: if boot failed, say so in chat and retry once
+    try:
+        llm = _load()
+    except Exception as e:
+        err = _boot_error or f"{type(e).__name__}: {e}"
+        yield history + [
+            {"role": "user", "content": message},
+            {"role": "assistant", "content": f"⚠️ Model failed to load: `{err}`\n\nMost likely: hardware too small (needs A10G or 32GB CPU space) or still downloading — try again in a minute."},
+        ]
+        return
+
     file_context = parse_all_files(uploaded_files)
     messages = [{"role": "system", "content": SYSTEM_PROMPT}]
 
@@ -168,18 +188,17 @@ def respond(message, history, uploaded_files):
     for entry in history:
         role = entry.get("role", "user")
         content = entry.get("content", "")
-        if content:
+        if isinstance(content, str) and content:
             messages.append({"role": role, "content": content})
 
     messages.append({"role": "user", "content": message})
 
-    llm = _load()
     stream = llm.create_chat_completion(
         messages=messages,
         temperature=0.6,
         top_p=0.8,
         top_k=20,
-        max_tokens=4096,
+        max_tokens=MAX_TOKENS,
         stream=True,
     )
 
@@ -195,13 +214,8 @@ def respond(message, history, uploaded_files):
             {"role": "assistant", "content": partial},
         ]
 
-    yield history + [
-        {"role": "user", "content": message},
-        {"role": "assistant", "content": partial},
-    ]
-
 def create_demo():
-    with gr.Blocks(title="codeMax — Qwen 3.6 27B Coder") as demo:
+    with gr.Blocks(title="codeMax — Qwen 3.6 27B Coder", css=CSS) as demo:  # FIX: css here, not in launch()
         gr.Markdown(
             "# codeMax\n"
             "**Qwen 3.6 27B · Uncensored · Code-optimized · Dedicated GPU**\n"
@@ -213,6 +227,7 @@ def create_demo():
                 chatbot = gr.Chatbot(
                     label="Chat",
                     height=580,
+                    type="messages",  # FIX: explicit, matches respond() format
                     avatar_images=(None, "https://huggingface.co/front/assets/huggingface_logo-noborder.svg"),
                 )
                 with gr.Row():
@@ -264,5 +279,4 @@ if __name__ == "__main__":
         server_name="0.0.0.0",
         server_port=7860,
         ssr_mode=False,
-        css="""footer { display: none !important; }""",
-    )
+    )  # FIX: css removed from launch()