Kosmos 2.5

[conceptofmind]

Hello,

Thank you for all your great research.

I was wondering if there were plans to release Kosmos 2.5 similar to how Kosmos 2 was released on Huggingface.

Thank you,

Enrico

[pribadihcr]

+1

[logan-markewich]

+1

[FrancescoSaverioZuppichini]

+1

[fernando-neto-ai]

+1

[evolu8]

please +1

[luvwinnie]

+1

[shenguojun]

+1

[MiaSanLei]

+1

[dydxdt]

+1

[coder4nlp]

+1

[zykzyk1993]

+1 please

[ydshieh]

+1 please 🙏

[acul3]

• 1

[contrebande-labs]

Would be nice if there was an official acknowledgement of this.

[Natyren]

Hi @wolfshow, сould you please let people there know if there are plans to release the code and weights for this model?

[wolfshow]

Thank you all for expressing interest in Kosmos-2.5. We plan to release the code and models, though we currently do not have a specific ETA. It's important to note that we are currently facing limitations in GPU resources. Despite this challenge, we are committed to making the models extremely user-friendly. There is still a significant amount of work ahead of us before we can open source the project, and we appreciate your understanding and patience as we navigate these constraints. We look forward to updating you on our progress.

[ydshieh]

Thank you @wolfshow for the information. Super glad your team considers to open source this exciting model!

[FrancescoSaverioZuppichini]

Thank you all for expressing interest in Kosmos-2.5. We plan to release the code and models, though we currently do not have a specific ETA. It's important to note that we are currently facing limitations in GPU resources. Despite this challenge, we are committed to making the models extremely user-friendly. There is still a significant amount of work ahead of us before we can open source the project, and we appreciate your understanding and patience as we navigate these constraints. We look forward to updating you on our progress.

Thanks a lot @wolfshow for breaking the silence :) I have to wonder, the paper did shows result that result must have come from some code and some trained model. Why do you need to use GPUs now? Would it be possible to release what you have done for the paper and let the community take care of the rest? Otherwise, it is very hard for us to validate the results and also to take advantage of the research done. I am sure all the authors would love to see people using their work.

Thank you again

[logan-markewich]

@FrancescoSaverioZuppichini it's microsoft research, so I'm sure they have their own red tape, responsibilities, and procedures to adhere too :)

[FrancescoSaverioZuppichini]

@FrancescoSaverioZuppichini it's microsoft research, so I'm sure they have their own red tape, responsibilities, and procedures to adhere too :)

Sure, They could have said it weeks ago if they cared to share them :) Well looks like this discussion as come to an end.

TL;DR At some point hopefully when the model is still relevant, we'll get some code that for some reason needs GPUs time while the paper showed results that must have been run on GPUs. We are living in the matrix

[wolfshow]

@FrancescoSaverioZuppichini it's microsoft research, so I'm sure they have their own red tape, responsibilities, and procedures to adhere too :)

Sure, They could have said it weeks ago if they cared to share them :) Well looks like this discussion as come to an end.

TL;DR At some point hopefully when the model is still relevant, we'll get some code that for some reason needs GPUs time while the paper showed results that must have been run on GPUs. We are living in the matrix

We need computations to run more experiements which are not included in the paper right now.

[evolu8]

Thank you @wolfshow . Your efforts are very much appreciated! Good luck with tying off the final pieces.

[shubhamagarwal92]

Hi @wolfshow! Interesting work on Kosmos 2.5!

Is there any plan to release the datasets as well? Or the corresponding pipeline code for different datasets (Sec 2.3/2.4 in the paper)?

[evolu8]

@wolfshow - any news?

[evolu8]

@wolfshow - another month, another nudge :) Please keep us posted.

[ydshieh]

Really hope this could be released - it would have huge impact 🙏 🔥

[coder4nlp]

Please!

[FrancescoSaverioZuppichini]

Hard F

[atlury]

Any ETA for the code and models for this paper? Its been nearly 3-4 months.

[radiachkik]

+1

[luohao123]

Still plan to release the code? The opensource community really needs such a work!

[sairin94]

+1

[AlexAndrei98]

+11 🚀 🚀 🚀

[wolfshow]

Kosmos 2.5 models and code have been publicly available at https://huggingface.co/microsoft/kosmos-2.5 and https://github.com/microsoft/unilm/tree/master/kosmos-2.5. Thanks!

[wolfshow]

Hello,

Thank you for all your great research.

I was wondering if there were plans to release Kosmos 2.5 similar to how Kosmos 2 was released on Huggingface.

Thank you,

Enrico

https://github.com/huggingface/transformers/issues/30877

[EwoutH]

Maybe notable: Kosmos-2.5 now also is published in Safetensors format on HuggingFace.

[atlury]

Ok an example with image input on hugging spaces would be highly helpful.