Kaushik Srivatsan

icon indicating an email [email protected]

icon location India icon location B.Tech in Computing & Data Science | Aspiring ML Researcher & Developer | ex-Intern eYantra, IIT Bombay

Results 3 comments of Kaushik Srivatsan

Device map feature for maestro models -qwen_2.5, florence_2 & paligemma_2

Hey @SkalskiP, The main reason I dropped the device argument completely is that I felt having two arguments that dealt with handling devices `device` and `device_map` might confuse the user...

Potential issue with load_modal function from trainer chekpoints for all models

@SkalskiP Thank you for following up on the issue really quickly. I think the solution you proposed broadly will work along with the one by @Matvezy to resolve this issue....

Potential issue with load_modal function from trainer chekpoints for all models

@SkalskiP Yes will be happy to. On a note unrelated to this, i was wondering whether it would be good if the load model can take in flash attention as...