Optimize referee model to Qwen2.5-0.5B-Instruct for CPU and add warning popup

Dockerfile

@@ -17,17 +17,13 @@ RUN pip install --no-cache-dir .
 RUN mkdir -p /app/nltk_data && \
     python -c "import nltk; nltk.download('punkt', download_dir='/app/nltk_data'); nltk.download('punkt_tab', download_dir='/app/nltk_data'); nltk.download('stopwords', download_dir='/app/nltk_data')"
 
-# Pre-download public HF embedding models
+# Pre-download public HF embedding and LLM models
 RUN python -c "from huggingface_hub import snapshot_download; \
     snapshot_download(repo_id='nomic-ai/nomic-embed-text-v1.5'); \
-    snapshot_download(repo_id='C10X/Qwen3-Embedding-TurboX.v2')"
+    snapshot_download(repo_id='C10X/Qwen3-Embedding-TurboX.v2'); \
+    snapshot_download(repo_id='Qwen/Qwen2.5-0.5B-Instruct')"
 
 
-# Download lightweight GGUF model instead of raw safetensors
-RUN mkdir -p /app/models && \
-    python -c "from huggingface_hub import hf_hub_download; \
-    hf_hub_download(repo_id='Qwen/Qwen2.5-1.5B-Instruct-GGUF', filename='qwen2.5-1.5b-instruct-q4_k_m.gguf', local_dir='/app/models')"
-
 COPY . .
 
 RUN chmod -R 777 /app

backend/engines/llm_client.py

@@ -7,7 +7,7 @@ _llm_pipeline = None
 
 class OllamaClient:
     def __init__(self):
-        self.model_name = "Qwen/Qwen2.5-1.5B-Instruct"
+        self.model_name = "Qwen/Qwen2.5-0.5B-Instruct"
 
     def _get_pipeline(self):
         global _llm_pipeline

frontend/app.js

@@ -1676,6 +1676,8 @@ if (domReferee.btnCall) {
     const labelA = domArena.modelA.options[domArena.modelA.selectedIndex].text;
     const labelB = domArena.modelB.options[domArena.modelB.selectedIndex].text;
 
+    showToast("Summoning AI referee. Running Qwen2.5-0.5B-Instruct on CPU container — evaluation will take ~15-20 seconds.", "info");
+
     domReferee.triggerBox.style.display = "none";
     domReferee.loadingBox.style.display = "block";
     domReferee.verdictBox.style.display = "none";