[Doc][Train] Add accelerator_type to Ray Train user guide
#44882
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Signed-off-by: Hongpeng Guo <hpguo@anyscale.com>
hongpeng-guo
requested review from
matthewdeng,
justinvyu,
woshiyyya and
a team
as code owners
|
Tip if you haven't seen this already: we build the docs as part of the premerge CI, so you can take a look at your rendered docs: https://anyscale-ray--44882.com.readthedocs.build/en/44882/index.html 
|
Nice tips! ty |
justinvyu
reviewed
Apr 22, 2024
Comment on lines
109
to
111
| Sometimes you might want to specify the accelerator type for a worker. For example, | ||
| you can specify `accelerator_type="A100"` in the `ScalingConfig` if you want to | ||
| assign the worker an NVIDIA A100 GPU. |
There was a problem hiding this comment.
Suggested change
| Sometimes you might want to specify the accelerator type for a worker. For example, | |
| you can specify `accelerator_type="A100"` in the `ScalingConfig` if you want to | |
| assign the worker an NVIDIA A100 GPU. | |
| Ray Train allows you to specify the accelerator type for each worker. | |
| This is useful if your model training has some GPU memory constraints that requires a specific type of GPU. | |
| In a heterogeneous Ray cluster, this means that your training workers will be forced to run on the specified GPU type, rather than on any arbitrary GPU node. | |
| For example, you can specify `accelerator_type="A100"` in the :class:`~ray.train.ScalingConfig` if you want to | |
| assign each worker a NVIDIA A100 GPU. |
Comment on lines
119
to
138
| import torch | ||
| from ray.train import ScalingConfig | ||
| from ray.train.torch import TorchTrainer, get_device | ||
|
|
||
|
|
||
| def train_func(): | ||
| assert torch.cuda.is_available() | ||
|
|
||
| device = get_device() | ||
| assert device == torch.device("cuda:0") | ||
|
|
||
| trainer = TorchTrainer( | ||
| train_func, | ||
| scaling_config=ScalingConfig( | ||
| num_workers=1, | ||
| use_gpu=True, | ||
| accelerator_type="A100" | ||
| ) | ||
| ) | ||
| trainer.fit() |
There was a problem hiding this comment.
We can cut this down to just show the ScalingConfig.
justinvyu
reviewed
Apr 22, 2024
Comment on lines
121
to
132
| from ray.train import ScalingConfig | ||
| from ray.train.torch import TorchTrainer | ||
|
|
||
|
|
||
| trainer = TorchTrainer( | ||
| train_func, | ||
| scaling_config=ScalingConfig( | ||
| num_workers=1, | ||
| use_gpu=True, | ||
| accelerator_type="A100" | ||
| ) | ||
| ) |
There was a problem hiding this comment.
Oh, for this one, I'm thinking of just showing:
ScalingConfig(...)
Comment on lines
116
to
117
| Ensure that your cluster has instances with the specified accelerator type | ||
| or is able to autoscale to fulfill the request. |
There was a problem hiding this comment.
We can make this a tip:
.. tip::
Ensure that your cluster has instances with the specified accelerator type
or is able to autoscale to fulfill the request.
Otherwise, your job will hang forever due to unsatisfiable pending resource requests.
There was a problem hiding this comment.
Oh nice tip structure. Will try.
Signed-off-by: Hongpeng Guo <hpguo@anyscale.com>
…accelerator-type
Signed-off-by: Hongpeng Guo <hpguo@anyscale.com>
woshiyyya
reviewed
Apr 22, 2024
Comment on lines
123
to
127
| ScalingConfig( | ||
| num_workers=1, | ||
| use_gpu=True, | ||
| accelerator_type="A100" | ||
| ) |
| Setting the GPU type | ||
| ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ | ||
| Ray Train allows you to specify the accelerator type for each worker. | ||
| This is useful if your model training has some GPU memory constraints that requires a specific type of GPU. |
There was a problem hiding this comment.
Users may want to use different accelerator types not only for GPU memory constraints, but also for e.g. compute power, cost efficiency, availability, etc.
Let's just say This is useful if you want to use a specific accelerator type for model training.
Signed-off-by: Hongpeng Guo <hpguo@anyscale.com>
Signed-off-by: Hongpeng Guo <hpguo@anyscale.com>
Signed-off-by: Hongpeng Guo <hpguo@anyscale.com>
justinvyu
changed the title
accelerator_type to Ray Train user guidesaccelerator_type to Ray Train user guide
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Why are these changes needed?
Our
ScalingConfig()function supports a new argumentaccelerator_type. This PR provides a user guide with example code to showcase the usage. The generated section of the user guide is appended below:Related issue number
"Closes #44763
Checks
git commit -s) in this PR.scripts/format.shto lint the changes in this PR.method in Tune, I've added it in
doc/source/tune/api/under thecorresponding
.rstfile.