Fix values in quantize_config.json

I noticed the values in `quantize_config.json` don't match the settings listed for the main branch in the model table in the README. After pulling the latest changes from main, I get CUDA illegal memory access errors during generation. These changes allow me to run the model again.

The only other model I checked was `guanaco-65B-GPTQ`, but it seems like it might be in a similar situation.

quantize_config.json

@@ -1,8 +1,8 @@
 {
     "bits": 4,
-    "group_size": 128,
+    "group_size": -1,
     "damp_percent": 0.01,
-    "desc_act": false,
+    "desc_act": true,
     "sym": true,
     "true_sequential": true
 }