[gemini] fix init bugs for modules (#2047)

* [gemini] fix init bugs for modules

* fix bugs

colossalai/utils/model/colo_init_context.py

@@ -96,10 +96,6 @@ class ColoInitContext(InsertPostInitMethodToModuleSubClasses):
         The function to call at the end of the constructor of each module.
         FIXME(fjr) The module may be passed to this function multiple times?
         """
-
-        if hasattr(module, '_colo_visited'):
-            return
-
         name_list = []
         for name, param in _named_params_with_replica(module):
             if isinstance(param, ColoTensor):
@@ -130,7 +126,6 @@ class ColoInitContext(InsertPostInitMethodToModuleSubClasses):
             colo_param.shared_param_modules.append(submodule)
 
         module.to(self._device)
-        ColoModulize(module)
 
 
 def post_process_colo_init_ctx(model: torch.nn.Module,

tests/test_gemini/update/test_optim.py

@@ -24,6 +24,11 @@ from tests.components_to_test import run_fwd_bwd
 from tests.components_to_test.registry import non_distributed_component_funcs
 from tests.test_tensor.common_utils import debug_print, set_seed
 
+# this model is large enough to slice to chunks
+TEST_MODELS = ['gpt2']
+# these models are too small, all parameters in these models are compacted into one chunk
+EXAMPLE_MODELS = ['hanging_param_model', 'bert', 'simple_net', 'nested_model', 'repeated_computed_layers']
+
 
 def check_param(model: ZeroDDP, torch_model: torch.nn.Module):
     zero_dict = model.state_dict(only_rank_0=False)
@@ -40,10 +45,6 @@ def check_param(model: ZeroDDP, torch_model: torch.nn.Module):
         assert_close(value, temp_zero_value, rtol=1e-3, atol=1e-2)
 
 
-# 'gpt2', 'bert',
-TEST_MODELS = ['hanging_param_model', 'gpt2', 'bert', 'simple_net', 'nested_model', 'repeated_computed_layers']
-
-
 @parameterize('placement_policy', ['cuda', 'cpu', 'auto', 'const'])
 @parameterize('model_name', TEST_MODELS)
 def exam_model_step(placement_policy, model_name: str):
@@ -61,8 +62,6 @@ def exam_model_step(placement_policy, model_name: str):
     with ColoInitContext(device=init_dev):
         model = model_builder()
 
-    post_process_colo_init_ctx(model, device=init_dev)
-
     for torch_p, p in zip(torch_model.parameters(), model.parameters()):
         p.data.copy_(torch_p.data)
 
@@ -102,8 +101,8 @@ def exam_model_step(placement_policy, model_name: str):
         check_param(model, torch_model)
 
 
-@parameterize('placement_policy', ['cuda', 'cpu'])
+@parameterize('placement_policy', ['cuda', 'cpu', 'auto', 'const'])
-@parameterize('model_name', TEST_MODELS)
+@parameterize('model_name', EXAMPLE_MODELS)
 def exam_tiny_example(placement_policy, model_name: str):
     set_seed(2008)
     get_components_func = non_distributed_component_funcs.get_callable(model_name)
@@ -119,8 +118,6 @@ def exam_tiny_example(placement_policy, model_name: str):
     with ColoInitContext(device=init_dev):
         model = model_builder()
 
-    post_process_colo_init_ctx(model, device=init_dev)
-
     for torch_p, p in zip(torch_model.parameters(), model.parameters()):
         p.data.copy_(torch_p.data)