Update GPT Fast handler to add option to get GPU ID from context #2872
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
lxning
reviewed
Jan 2, 2024
Comment on lines
53
to
59
| if ( | ||
| "torchrun" not in ctx.model_yaml_config | ||
| and properties.get("gpu_id") is not None | ||
| ): | ||
| gpu_id = properties.get("gpu_id") | ||
| else: | ||
| gpu_id = self.local_rank |
There was a problem hiding this comment.
this part is not necessary.
There was a problem hiding this comment.
Understood. Will update in subsequent commit.
Comment on lines
61
to
63
| self.device = torch.device(self.map_location + ":" + str(gpu_id)) | ||
|
|
||
| torch.cuda.set_device(self.local_rank) | ||
| torch.cuda.set_device(gpu_id) |
There was a problem hiding this comment.
Understood. Will update in subsequent commit
| @@ -44,15 +44,23 @@ def __init__(self): | |||
|
|
|||
| def initialize(self, ctx): | |||
| self.context = ctx | |||
| properties = self.context.system_properties | |||
| rank = maybe_init_dist() | |||
|
|
|||
| self.local_rank = rank if rank is not None else 0 | |||
There was a problem hiding this comment.
if rank is none, local rank should be equal to device_id on gpu host.
There was a problem hiding this comment.
Thanks for the suggestion Li. Are you suggesting that we implement the handler in this way?
def initialize(self, ctx):
self.context = ctx
properties = self.context.system_properties
gpu_id = properties.get("gpu_id")
rank = maybe_init_dist()
self.local_rank = rank if rank is not None else int(gpu_id)
if torch.cuda.is_available():
self.map_location = "cuda"
self.device = torch.device(self.map_location + ":" + str(self.local_rank))
torch.cuda.set_device(self.local_rank)
There was a problem hiding this comment.
please add "raise exception" when gpu_id < 0
There was a problem hiding this comment.
Understood. I will include the exception
lxning
approved these changes
Jan 3, 2024
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.
Description
Please read our CONTRIBUTING.md prior to creating your first pull request.
Creating PR to update the handler for the GPT Fast example to support obtaining the GPU ID from context.
Testing:
Created two model artifacts i.e.
gpt_fast_7b.mar(4 workers without tensor parallelism) andgpt_fast_7b_tp.mar(1 worker with tensor parallelism and parallel level 4) with modified handler. Performed inference with both models:Case 1:
gpt_fast_7b.marOutput:
nvidia-smioutput:Tar file with model artifact MAR file and TorchServe logs: gpt_fast_7b.tar.gz
Case 2:
gpt_fast_7b_tp.marOutput:
nvidia-smioutput:Tar file with model artifact MAR file and TorchServe logs: gpt_fast_7b_tp.tar.gz