Commit graph

5 commits

Author SHA1 Message Date
Jiajia Qin
7e0dd9d433
[js/webgpu] Optimize Expand (#22752)
Use components = 4 if possible.

llama3.2-1B becomes 20 tokens/s from 18 tokens/s on my iGPUs.
2024-11-12 12:37:19 -08:00
Jiajie Hu
447a3a7c70
[js/webgpu] Fix Expand/Gather when input type is bool (#18999)
### Description
Also update the op test suite.

### Motivation and Context
Previously the *total* size in case `Expand - last dim is not divisible
by 4` was a multiple of 4, even though the *last dimension* was not, so
the bug has never been caught.
2024-01-05 08:16:15 -08:00
Jiajia Qin
6781b6cf3d
[js/webgpu] add bool type for Expand/Gather (#18615)
### Description
In [detr-resnet-50](https://huggingface.co/Xenova/detr-resnet-50) model,
it uses expand with bool type running on cpu ep.




| Kernel    | Shape | Provider |
| -------- | ------- | ------- |
| Expand | "input_type_shape" :
[{"bool":[1,1,1,625]},{"int64":[4]}],"activation_size" :
"657","output_type_shape" : [{"bool":[1,1,625,625]}] |
CPUExecutionProvider |

After this change, it will run on jsep.
| Kernel    | Shape | Provider |
| -------- | ------- | ------- |
| Expand | "input_type_shape" :
[{"bool":[1,1,1,625]},{"int64":[4]}],"activation_size" :
"657","output_type_shape" : [{"bool":[1,1,625,625]}] |
JsExecutionProvider |
2023-11-30 15:47:08 -08:00
satyajandhyala
841f7ed3e0
[[JS/Web]Added uniform to Expand op. (#18558)
### Description
<!-- Describe your changes. -->
Added Uniforms to Expand operator kernel


### Motivation and Context
<!-- - Why is this change required? What problem does it solve?
- If it fixes an open issue, please link to the issue here. -->
Improve performance
2023-11-22 14:14:24 -08:00
Hariharan Seshadri
66df11769c
[JS/WebGPU] Expand operator fixes (#17137) 2023-08-16 11:24:26 -07:00