MCPcopy Create free account
hub / github.com/deepspeedai/DeepSpeed / create_deepspeed_model

Function create_deepspeed_model

tests/unit/checkpoint/common.py:157–163  ·  view source on GitHub ↗
(config_dict, model, base_optimizer)

Source from the content-addressed store, hash-verified

155
156
157def create_deepspeed_model(config_dict, model, base_optimizer):
158 ds_model, _, _, _ = deepspeed.initialize(config=config_dict,
159 model=model,
160 model_parameters=create_moe_param_groups(model),
161 optimizer=base_optimizer)
162 ds_model.empty_partition_cache()
163 return ds_model
164
165
166def checkpoint_correctness_verification(config_dict,

Calls 3

create_moe_param_groupsFunction · 0.85
initializeMethod · 0.80
empty_partition_cacheMethod · 0.45

Tested by 6

runMethod · 0.68