Method bfloat16_enabled

deepspeed/runtime/engine.py:1314–1315 · view source on GitHub ↗

(self)

Source from the content-addressed store, hash-verified

1312	return self._config.float16_config.enabled
1313
1314	def bfloat16_enabled(self):
1315	return self._config.bfloat16_config.enabled
1316
1317	def fp16_master_weights_and_gradients(self):
1318	return self._config.float16_config.fp16_master_weights_and_grads

__init__Method · 0.95

communication_data_typeMethod · 0.95

get_data_typesMethod · 0.95

_assert_valid_mixed_precision_configMethod · 0.95

_mixed_precision_dtypesMethod · 0.95

_do_sanity_checkMethod · 0.95

_configure_distributed_modelMethod · 0.95

_configure_zero_optimizerMethod · 0.95

_take_model_stepMethod · 0.95

_get_zero_ckpt_nameMethod · 0.95

_load_checkpointMethod · 0.95

_get_all_zero_checkpointsMethod · 0.95

no outgoing calls

engine_input_dtypeFunction · 0.64

test_grad_clip_and_norm_updateMethod · 0.64