Update LLM model configuration

app.py

@@ -13,8 +13,8 @@ from llama_cpp import Llama
 
 TTS_MODEL_ID = os.getenv("TTS_MODEL_ID", "Qwen/Qwen3-TTS-12Hz-0.6B-CustomVoice")
 
-LLM_REPO_ID = os.getenv("LLM_REPO_ID", "Qwen/Qwen2.5-1.5B-Instruct-GGUF")
-LLM_FILENAME = os.getenv("LLM_FILENAME", "qwen2.5-1.5b-instruct-q4_k_m.gguf")
+LLM_REPO_ID = os.getenv("LLM_REPO_ID", "Qwen/Qwen2.5-0.5B-Instruct-GGUF")
+LLM_FILENAME = os.getenv("LLM_FILENAME", "qwen2.5-0.5b-instruct-q4_k_m.gguf")
 
 tts_model = None
 llm_model = None
@@ -146,7 +146,7 @@ def get_llm_model():
     llm_model = Llama.from_pretrained(
         repo_id=LLM_REPO_ID,
         filename=LLM_FILENAME,
-        n_ctx=2048,
+        n_ctx=1024,
         n_threads=max(2, os.cpu_count() or 2),
         n_gpu_layers=-1 if torch.cuda.is_available() else 0,
         verbose=True,
@@ -195,7 +195,7 @@ Rules:
             {"role": "user", "content": user_prompt},
         ],
         temperature=0.2,
-        max_tokens=256,
+        max_tokens=96,
     )
 
     content = result["choices"][0]["message"]["content"].strip()
@@ -271,10 +271,10 @@ with gr.Blocks(
             Current milestone: llama.cpp intent generation + Qwen3-TTS voice synthesis.
           </p>
           <div class="badge-row">
-            <span class="badge">🦙 llama.cpp</span>
-            <span class="badge">🔌 local-first</span>
-            <span class="badge">🎨 custom Gradio UI</span>
-            <span class="badge">🗣️ expressive TTS</span>
+            <span class="badge">llama.cpp</span>
+            <span class="badge">local-first</span>
+            <span class="badge">custom Gradio UI</span>
+            <span class="badge">expressive TTS</span>
           </div>
         </section>
         """