FP32/FP16 to BF16 Model Conversion: figures difficult to read and understand

[bconsolvo]

Looking here:
https://quark.docs.amd.com/latest/supported_accelerators/ryzenai/tutorial_convert_fp32_or_fp16_to_bf16.html

Figures 1 and 2 are very difficult to read and understand. Is there a better way to represent these?

[ChaoLi-AMD]

Hi @bconsolvo, thanks for raising this.

Figures 1 and 2 are accompanied by textual descriptions, and the images can be clicked to view them at a larger size. Could you clarify what specifically is difficult to read or understand?

Also, could you elaborate on what you mean by “a better way to represent these”? If you have a preferred format, we would be happy to consider it.

[bconsolvo]

Even after clicking into Figure 2, very difficult to read the text on each component. For Figure 2, I recommend splitting this into multiple figures that are higher quality, so you can actually read the text.

Figure 1: I would remove this entirely. Can be described in text, and the diagram is not helping me to understand anything further than what the text describes.

[ChaoLi-AMD]

Thank you @bconsolvo for the detailed feedback, your suggestions make sense. We’ll revisit this section and improve the figures in a future documentation update. We appreciate your input!

[larryli2-amd]

Hi @bconsolvo
Thank you for your commnt. We’ll release Quark 0.13 in about a month, and at that point you’ll see that this page has been updated: Quark BF16 conversion documentation.
cc @ChaoLi-AMD