Accelerator: HPU training¶

Audience: Gaudi chip users looking to use advanced strategies and profiling HPU’s.

Working with HPUProfiler¶

HPUProfiler is a lightning implementation of PyTorch profiler for HPU devices. It aids in obtaining profiling summary of PyTorch functions. It subclasses PyTorch Lightning’s PyTorch profiler.

Default Profiling¶

For auto profiling, create a HPUProfiler instance and pass it to trainer. At the end of profiler.fit(), it will generate a json trace for the run. In case accelerator= HPUAccelerator() is not used with HPUProfiler, then it will dump only CPU traces, similar to PyTorchProfiler.

from lightning import Trainer
from lightning_habana.pytorch.accelerator import HPUAccelerator
from lightning_habana.pytorch.profiler.profiler import HPUProfiler

trainer = Trainer(accelerator=HPUAccelerator(), profiler=HPUProfiler())

Distributed Profiling¶

To profile a distributed model, use the HPUProfiler with the filename argument which will save a report per rank:

from pytorch_lightning import Trainer
from lightning_habana.pytorch.accelerator import HPUAccelerator
from lightning_habana.pytorch.profiler.profiler import HPUProfiler

profiler = HPUProfiler(filename="perf-logs")
trainer = Trainer(profiler=profiler, accelerator=HPUAccelerator())

Custom Profiling¶

To profile custom actions of interest, reference a profiler in the LightningModule.:

from pytorch_lightning import Trainer
from lightning_habana.pytorch.accelerator import HPUAccelerator
from lightning_habana.pytorch.profiler.profiler import HPUProfiler

# Reference profiler in LightningModule
class MyModel(LightningModule):
    def __init__(self, profiler=None):
        self.profiler = profiler

# To profile in any part of your code, use the self.profiler.profile() function
    def custom_processing_step_basic(self, data):
        with self.profiler.profile("my_custom_action"):
            print("do somthing")
        return data

# Alternatively, use self.profiler.start("my_custom_action")
# and self.profiler.stop("my_custom_action") functions
# to enclose the part of code to be profiled.
    def custom_processing_step_granular(self, data):
        self.profiler.start("my_custom_action")
        print("do somthing")
        self.profiler.stop("my_custom_action")
        return data

# Pass profiler instance to LightningModule
profiler = HPUProfiler()
model = MyModel(profiler)
trainer = Trainer(accelerator=HPUAccelerator(), profiler=profiler)

For more details on profiler, refer to PyTorchProfiler

Visualize Profiled Operations¶

Profiler will dump traces in json format. The traces can be visualized in 2 ways:

Using PyTorch TensorBoard Profiler¶

For further instructions see, https://github.com/pytorch/kineto/tree/master/tb_plugin.

Install tensorboard¶

python -um pip install tensorboard torch-tb-profiler

Start the TensorBoard server (default at port 6006)¶

tensorboard --logdir ./tensorboard --port 6006

Now open the following url in your browser¶

http://localhost:6006/#profile

Using Chrome¶

Open Chrome and copy/paste this URL: chrome://tracing/.

Once tracing opens, click on Load at the top-right and load one of the generated traces.

Limitations¶

When using the HPUProfiler, wall clock time will not be representative of the true wall clock time. This is due to forcing profiled operations to be measured synchronously, when many HPU ops happen asynchronously. It is recommended to use this Profiler to find bottlenecks/breakdowns, however for end to end wall clock time use the SimpleProfiler.
HPUProfiler.summary() is not supported
Passing profiler name as string “hpu” to the trainer is not supported.

Working with DeepSpeed¶

HPU’s support advanced strategies like deepspeed. By default, HPU training will use 32-bit precision. To enable mixed precision, set the precision flag.

from lightning.pytorch.plugins import DeepSpeedPrecisionPlugin
from lightning_habana.pytorch.accelerator import HPUAccelerator
from lightning_habana.pytorch.strategies import HPUDeepSpeedStrategy

trainer = Trainer(devices=8, accelerator=HPUAccelerator(), strategy=HPUDeepSpeedStrategy(), plugins=[DeepSpeedPrecisionPlugin(precision="bf16-mixed")])

More details on the HPU supported deepspeed features and functionalities, refer to refer to Deepspeed with HPU