Lower Gemma AWQ group size to 16

quantize_to_awq_colab.ipynb

@@ -112,7 +112,7 @@
    "source": [
     "# Model-specific AWQ overrides. Keys match MODELS_TO_QUANTIZE entries.\n",
     "MODEL_AWQ_OVERRIDES = {\n",
-    "    \"router-gemma3-merged\": {\"group_size\": 64},\n",
+    "    \"router-gemma3-merged\": {\"group_size\": 16},\n",
     "}\n",
     "\n",
     "# Derived AWQ configs per model (defaults + overrides)\n",
@@ -129,16 +129,6 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "# Model-specific overrides for AWQ quantization\n",
-    "# These are merged into the base AWQ config inside quantize_model_to_awq\n",
-    "MODEL_AWQ_OVERRIDES = {\n",
-    "    # Gemma 3 uses linear layers whose column widths are not divisible by 128.\n",
-    "    # Using group_size=64 avoids quantization failures while retaining accuracy.\n",
-    "    \"gemma\": {\n",
-    "        \"group_size\": 64,\n",
-    "    },\n",
-    "    # Add additional overrides keyed by model_type as needed\n",
-    "}\n",
     "\n"
    ]
   },