Skip Triton import for AMD #5110
Merged
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
mrwyattii
reviewed
Feb 9, 2024
mrwyattii
approved these changes
Feb 9, 2024
mauryaavinash95
pushed a commit
to mauryaavinash95/DeepSpeed
that referenced
this pull request
Feb 17, 2024
When testing DeepSpeed inference on an `AMD Instinct MI250X/MI250` GPU, the `pytorch-triton-rocm` module would break the `torch.cuda` device API. To address this, importing `triton` is skipped when the GPU is determined to be `AMD`. This change allows DeepSpeed to be executed on an AMD GPU w/o kernel injection in the DeepSpeedExamples [text-generation example](https://github.com/microsoft/DeepSpeedExamples/tree/master/inference/huggingface/text-generation) using the following command: ```bash deepspeed --num_gpus 1 inference-test.py --model facebook/opt-125m ``` TODO: Root-cause the interaction between `pytorch-triton-rocm` and DeepSpeed to understand why this is causing the `torch.cuda` device API to break.
rraminen
pushed a commit
to ROCm/DeepSpeed
that referenced
this pull request
May 9, 2024
When testing DeepSpeed inference on an `AMD Instinct MI250X/MI250` GPU, the `pytorch-triton-rocm` module would break the `torch.cuda` device API. To address this, importing `triton` is skipped when the GPU is determined to be `AMD`. This change allows DeepSpeed to be executed on an AMD GPU w/o kernel injection in the DeepSpeedExamples [text-generation example](https://github.com/microsoft/DeepSpeedExamples/tree/master/inference/huggingface/text-generation) using the following command: ```bash deepspeed --num_gpus 1 inference-test.py --model facebook/opt-125m ``` TODO: Root-cause the interaction between `pytorch-triton-rocm` and DeepSpeed to understand why this is causing the `torch.cuda` device API to break.
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.
When testing DeepSpeed inference on an
AMD Instinct MI250X/MI250GPU, thepytorch-triton-rocmmodule would break thetorch.cudadevice API. To address this, importingtritonis skipped when the GPU is determined to beAMD.This change allows DeepSpeed to be executed on an AMD GPU w/o kernel injection in the DeepSpeedExamples text-generation example using the following command:
TODO: Root-cause the interaction between
pytorch-triton-rocmand DeepSpeed to understand why this is causing thetorch.cudadevice API to break.