Skip to content

Conversation

@Pratheesh-04-MCW
Copy link
Contributor

@Pratheesh-04-MCW Pratheesh-04-MCW commented Apr 9, 2025

This PR refactors the Stack Operation to be implemented at the core level using decomposed operations. It allows Stack to be accessed directly via R.Stack, making it more efficient and easier to use in TVM models. By adding this support we can run the following models

1. microsoft/beit-base-patch16-224-pt22k-ft22k (Image Classification)
2. anton-l/wav2vec2-random-tiny-classifier (Audio Classification)
3. microsoft/table-transformer-detection (Object Detection)
4. facebook/detr-resnet-50 (Object Detection)
5. facebook/mask2former-swin-tiny-coco-instance (Image Segmentation)
6. Intel/dpt-beit-large-512 (Depth Estimation)

@Pratheesh-04-MCW Pratheesh-04-MCW marked this pull request as ready for review April 17, 2025 08:21
@Pratheesh-04-MCW Pratheesh-04-MCW marked this pull request as draft April 17, 2025 08:21
@Pratheesh-04-MCW Pratheesh-04-MCW marked this pull request as ready for review April 17, 2025 08:22
@Pratheesh-04-MCW
Copy link
Contributor Author

@Hzfengsy Could you please review this PR?

@Hzfengsy Hzfengsy merged commit c00f52a into apache:main Apr 18, 2025
12 checks passed
ShiboXing pushed a commit to ShiboXing/tvm that referenced this pull request Aug 10, 2025
)

* add op support for stack

* trailing whitespace issue fixed

* fixed lint issues

* fixed whitespace issue

* fixed lint error

* fixing lint issues

* fixed whitespace issue

* add test script for fx_graph

* fix lint issues

* fixed unity check issues

* unity check

* fixed unity check issues

* lint issues
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants