Question about Text Embedding

[danielemolino]

Hi, I would like to understand the shape of the text embedding you use to condition the low resolution model.
I've tried your example and i've received a tensor of shape 1x192x768, why this choiche respect to like a 1x1x768?

[qianxingyucode]

hi, I really want to know that whether you get the pre-trained language model?

[kayhan-batmanghelich]

We used a pre trained model and fine tune it on 200k radiology reports. The details are in the paper. Sent from mobile phone. Sorry for misspellings and abbreviations.

…

On Fri, Mar 7, 2025 at 5:00 AM Qianxingyu ***@***.***> wrote: hi, I really want to know that whether you get the pre-trained language model? — Reply to this email directly, view it on GitHub <#5 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AC53JXIIWITL5AZK3NE4VIT2TFU25AVCNFSM6AAAAABYKDN7ASVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDOMBWGAZDKMJQHA> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***> [image: qianxingyucode]*qianxingyucode* left a comment (batmanlab/MedSyn#5) <#5 (comment)> hi, I really want to know that whether you get the pre-trained language model? — Reply to this email directly, view it on GitHub <#5 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AC53JXIIWITL5AZK3NE4VIT2TFU25AVCNFSM6AAAAABYKDN7ASVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDOMBWGAZDKMJQHA> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[danielemolino]

Can you also answer to me?

[danielemolino]

@qianxingyucode it is written in the ReadMe. Here it is! https://www.dropbox.com/scl/fi/d6tg6si72nnjfa87vawsl/pretrained_lm.gz?rlkey=fcnyrmy1i3xi9frzjchc68kh3&e=1&st=gq6xofnh&dl=0

[qianxingyucode]

https://www.dropbox.com/scl/fi/d6tg6si72nnjfa87vawsl/pretrained_lm.gz?rlkey=fcnyrmy1i3xi9frzjchc68kh3&e=1&st=gq6xofnh&dl=0

I notice this link, but it seem to be broken(I tried in different network, different devices), so I can't get the pre-trained model.
Could you please offer me a new link? Or send to my email: xingq004@gmail.com

Thank you for your reply! 😄

[lisun-ai]

Hi @danielemolino I think it's a design choice. If we use 1x1x768, we will lose a lot of information. If we extract longer sequences, we may encounter excessive GPU memory usage.
Hi @qianxingyucode, I just tried this link and it works. Unfortunately, Dropbox is blocked in China, so you may need to use VPN to access this site.