Update app.py

app.py

@@ -1,4 +1,4 @@
-import os
+mport os
 import gradio as gr
 import json
 import time
@@ -13,11 +13,25 @@ MODELS_DIR = "/data/models"
 MAX_TOKENS = 2048
 CONTEXT_SIZE = 4096
 
-# Map models to their likely GGUF repositories (Falling back to Qwen for the 3B)
 MODEL_REPOS = {
+    # 30B: Unsloth is the most reliable source for Qwen3 GGUFs currently
+    "Qwen3-Coder-30B-A3B-Instruct-Q4_K_M.gguf": "unsloth/Qwen3-Coder-30B-A3B-Instruct-GGUF",
+
+    # 3B: Qwen actually has an official one, but bartowski is safer fallback
     "qwen2.5-coder-3b-instruct-q4_k_m.gguf": "Qwen/Qwen2.5-Coder-3B-Instruct-GGUF",
-    "qwen2.5-coder-7b-instruct-q4_k_m.gguf": "Qwen/Qwen2.5-Coder-7B-Instruct-GGUF",
-    # You can add others here
+    
+    # 7B: Official Qwen GGUF is often missing/broken. Bartowski is the go-to here.
+    "qwen2.5-coder-7b-instruct-q4_k_m.gguf": "bartowski/Qwen2.5-Coder-7B-Instruct-GGUF",
+    
+    # 14B: Bartowski is recommended for consistency
+    "qwen2.5-coder-14b-instruct-q4_k_m.gguf": "bartowski/Qwen2.5-Coder-14B-Instruct-GGUF",
+    
+    # DeepSeek: Definitely needs community repo
+    "DeepSeek-Coder-V2-Lite-Instruct-Q4_K_M.gguf": "bartowski/DeepSeek-Coder-V2-Lite-Instruct-GGUF",
+    
+    # Tiny models
+    "qwen2.5-coder-1.5b-instruct-q4_k_m.gguf": "Qwen/Qwen2.5-Coder-1.5B-Instruct-GGUF",
+    "qwen2.5-coder-0.5b-instruct-q4_k_m.gguf": "Qwen/Qwen2.5-Coder-0.5B-Instruct-GGUF",
 }
 
 MODELS = {