Quantization of Other Models

[Oguzhanercan]

Hi, thanks for your work. I wonder that, why you named this as LLM and Diffusion model quantizer. I cannot see any problem about quantizing other models architectures with this like DINO, ViT etc. for other tasks (object detection, semantic segmentation). What is your insights about it?

[synxlin]

Hi,

There should not be any problems about quantizing other models.

We have LLM and Diffusion model quantizer just for adding specific quantization configurations separately.

[rohan-patankar]

is there any reference where i can convert the DINO model to onnx or other compression format? @synxlin @sxtyzhangzk @sxtyzhangzk it would be very helpful
thankyou..

[rohan-patankar]

Hi, thanks for your work. I wonder that, why you named this as LLM and Diffusion model quantizer. I cannot see any problem about quantizing other models architectures with this like DINO, ViT etc. for other tasks (object detection, semantic segmentation). What is your insights about it?

hi @Oguzhanercan can you pls help me i need to quantize the DINO model i have tried native approached like using direct torch module to convert to onnx but it failed and later i tried https://github.com/IDEA-Research/detrex this repo to convert to onnx after this my model latency got increased to 2x...may i pls know what approach have you dont to optimize the DINO model...it would be very helpful ....iam stuck in this past 1week..