only AUTO will try other quant linears (ModelCloud#797)

* only AUTO will try other linears

* add comments

* update comments

gptqmodel/utils/model.py

@@ -134,7 +134,10 @@ def make_quant(
         try:
             result = create_quant_layer(linear, bits, desc_act, dynamic, group_size, module, names, sym)
             return result
-        except NotImplementedError:
+        except NotImplementedError as e:
+            # only fallback to other quant linears when backend is auto.
+            if backend not in [BACKEND.AUTO, BACKEND.AUTO_TRAINABLE]:
+                raise e
             continue
 
     raise ValueError("no support quant linear was found for this module.")