[WIP] Support accelerate for GLM

modeling_glm.py

@@ -625,6 +625,7 @@ class GLMPreTrainedModel(PreTrainedModel):
     base_model_prefix = "glm"
     supports_gradient_checkpointing = True
     _keys_to_ignore_on_load_missing = [r"position_ids"]
+    _no_split_modules = ["GLMBlock"]
 
     def _init_weights(self, module):
         """ Initialize the weights """