diff --git a/tests/unit/runtime/half_precision/test_bf16.py b/tests/unit/runtime/half_precision/test_bf16.py index 740fa30641a1..3f551fb0fd4a 100644 --- a/tests/unit/runtime/half_precision/test_bf16.py +++ b/tests/unit/runtime/half_precision/test_bf16.py @@ -27,6 +27,7 @@ def test(self, zero_stage=2, use_cpu_offload=False): pytest.skip("cpu-adam is not compatible") config_dict = { + "train_micro_batch_size_per_gpu": 1, "steps_per_print": 1, "optimizer": { "type": "Adam", @@ -87,7 +88,7 @@ def test(self, zero_stage=2, use_cpu_offload=False): pytest.skip("cpu-adam is not compatible") config_dict = { - "train_batch_size": 4, + "train_micro_batch_size_per_gpu": 4, "steps_per_print": 1, "fp16": { "enabled": False, @@ -180,7 +181,7 @@ def test(self, optimizer_constructor, zero_stage=2): ) config_dict = { - "train_batch_size": 2, + "train_micro_batch_size_per_gpu": 2, "steps_per_print": 1, "fp16": { "enabled": False @@ -209,7 +210,7 @@ def test(self): ) config_dict = { - "train_batch_size": 2, + "train_micro_batch_size_per_gpu": 2, "steps_per_print": 1, "optimizer": { "type": "Adam", @@ -258,7 +259,7 @@ def test(self, stage=2): ) config_dict = { - "train_batch_size": 1, + "train_micro_batch_size_per_gpu": 1, "steps_per_print": 1, "fp16": { "enabled": False @@ -301,7 +302,7 @@ def test(self, comp_type, comm_type): type_str = {torch.float16: "fp16", torch.bfloat16: "bfp16"} config_dict = { - "train_batch_size": 2, + "train_micro_batch_size_per_gpu": 2, "steps_per_print": 1, "fp16": { "enabled": comp_type == torch.float16