Fix redundant shared_params in zero_to_fp32.py (#3149)

* submit changes

* update format

* fix fomrat

* revert

* test

* add top

* treat z1 as z2

* fix shared

* remove old changes

---------
Co-authored-by: NOlatunji Ruwase <olruwase@microsoft.com>

deepspeed/utils/zero_to_fp32.py

@@ -125,7 +125,7 @@ def parse_model_states(files):
         shared_params = []
         for param in state_dict["module"]:
             if param not in [*param_names, *buffer_names]:
-                for share_param in state_dict["module"]:
+                for share_param in [*param_names, *buffer_names]:
                     if (state_dict["module"][share_param].data_ptr() == state_dict["module"][param].data_ptr()
                             and share_param != param):
                         shared_params.append([param, share_param])
@@ -340,7 +340,8 @@ def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zer
 
     # recover shared parameters
     for pair in zero_model_states[0].shared_params:
-        state_dict[pair[0]] = state_dict[pair[1]]
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
 
     return state_dict
 
@@ -457,7 +458,8 @@ def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zer
 
     # recover shared parameters
     for pair in zero_model_states[0].shared_params:
-        state_dict[pair[0]] = state_dict[pair[1]]
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
 
     return state_dict