Update app.py

app.py

@@ -12,15 +12,25 @@ MAX_TOKENS = 2048
 CONTEXT_SIZE = 4096
 
 MODELS = {
-    "⭐ Qwen2.5 Coder 7B (Best)": "qwen2.5-coder-7b-instruct-q4_k_m.gguf",
-    "Qwen2.5 Coder 3B (Fast)": "qwen2.5-coder-3b-instruct-q4_k_m.gguf",
-    "Qwen2.5 Coder 1.5B (Fastest)": "qwen2.5-coder-1.5b-instruct-q4_k_m.gguf",
+    "⭐ Qwen3 Coder 30B-A3B (Best)": "Qwen3-Coder-30B-A3B-Instruct-Q4_K_M.gguf",
+    "🏆 Qwen2.5 Coder 14B (Premium)": "qwen2.5-coder-14b-instruct-q4_k_m.gguf",
+    "🧠 DeepSeek V2 Lite (Logic)": "DeepSeek-Coder-V2-Lite-Instruct-Q4_K_M.gguf",
+    "⚖️ Qwen2.5 Coder 7B (Balanced)": "qwen2.5-coder-7b-instruct-q4_k_m.gguf",
+    "🚀 Qwen2.5 Coder 3B (Fast)": "qwen2.5-coder-3b-instruct-q4_k_m.gguf",
+    "⚡ DeepSeek Coder 6.7B": "deepseek-coder-6.7b-instruct.Q4_K_M.gguf",
+    "💨 Qwen2.5 Coder 1.5B (Quick)": "qwen2.5-coder-1.5b-instruct-q4_k_m.gguf",
+    "🔬 Qwen2.5 Coder 0.5B (Instant)": "qwen2.5-coder-0.5b-instruct-q4_k_m.gguf",
 }
 
 MODEL_INFO = {
-    "⭐ Qwen2.5 Coder 7B (Best)": "🏆 Best quality • ~4.5GB",
-    "Qwen2.5 Coder 3B (Fast)": "⚖️ Balanced • ~2GB • Recommended",
-    "Qwen2.5 Coder 1.5B (Fastest)": "⚡ Fastest • ~1GB",
+    "⭐ Qwen3 Coder 30B-A3B (Best)": "🏆 Best quality • MoE 30B/3B • ~10GB",
+    "🏆 Qwen2.5 Coder 14B (Premium)": "💎 Premium • ~8GB • Complex tasks",
+    "🧠 DeepSeek V2 Lite (Logic)": "🧠 MoE 16B • ~9GB • Algorithms",
+    "⚖️ Qwen2.5 Coder 7B (Balanced)": "⚖️ Balanced • ~4.5GB • Recommended",
+    "🚀 Qwen2.5 Coder 3B (Fast)": "🚀 Fast • ~2GB • Great all-rounder",
+    "⚡ DeepSeek Coder 6.7B": "⚡ Logic focused • ~4GB",
+    "💨 Qwen2.5 Coder 1.5B (Quick)": "💨 Quick • ~1GB • Simple tasks",
+    "🔬 Qwen2.5 Coder 0.5B (Instant)": "🔬 Instant • ~0.3GB • Lightning fast",
 }
 
 LANGUAGES = [
@@ -118,10 +128,16 @@ def transcribe_audio(audio):
 def generate_response(model_name, prompt, temperature=0.7, max_tokens=2048):
     llm = load_model(model_name)
     if not llm:
-        return "❌ **Model not available.** Check if downloaded."
+        return "❌ **Model not available.**"
     
     try:
-        formatted = f"<|im_start|>system\nYou are an expert coding assistant.<|im_end|>\n<|im_start|>user\n{prompt}<|im_end|>\n<|im_start|>assistant\n"
+        # Different formats for different models
+        if "deepseek" in model_name.lower():
+            formatted = f"### Instruction:\n{prompt}\n\n### Response:\n"
+            stop_tokens = ["### Instruction:", "### Response:"]
+        else:  # Qwen format
+            formatted = f"<|im_start|>system\nYou are an expert coding assistant.<|im_end|>\n<|im_start|>user\n{prompt}<|im_end|>\n<|im_start|>assistant\n"
+            stop_tokens = ["<|im_end|>", "<|im_start|>"]
         
         output = llm(
             formatted,
@@ -130,7 +146,7 @@ def generate_response(model_name, prompt, temperature=0.7, max_tokens=2048):
             top_p=0.9,
             top_k=40,
             repeat_penalty=1.1,
-            stop=["<|im_end|>", "<|im_start|>"],
+            stop=stop_tokens,
             echo=False
         )
         
@@ -138,7 +154,6 @@ def generate_response(model_name, prompt, temperature=0.7, max_tokens=2048):
         return response if response else "⚠️ Empty response."
     except Exception as e:
         return f"❌ **Error:** {str(e)[:100]}"
-
 def extract_code(text):
     if not text or "```" not in text:
         return text