model conversion colab issue

[burntail]

https://colab.research.google.com/github/googlesamples/mediapipe/blob/main/examples/llm_inference/conversion/llm_conversion.ipynb?hl=ko#scrollTo=LSSxrLyQPofw

I tried to take converted models (ex) Falcon_cpu etc), but there is a issue like the picture which I attached.
Could explain about the issue and how to resolve the issue? I didn't change any other codes in colab.
Thank you.

[kuaashish]

Hi @burntail,

Thank you for reporting this issue. We are able to reproduce it; you can view the details here: Gist. Additionally, we have verified that other models on CPU are experiencing the same issue, while GPU conversion is working as expected. We have highlighted the issue, and our team is already working on a fix. We will provide updates as soon as we have more information.

[JakobPogacnikSouvent]

I am experiencing the same issue for CPU conversion of Falcon 1B, StableLM 3B and Phi 2 models. The GPU conversion of said models works as expected.

[kuaashish]

Hi @schmidt-sebastian,

We have reproduced the issue in the latest version (0.10.21). For your reference, we are attaching the Colab gist. Could you please review the issue again?

Thank you!!

[kuaashish]

Hi @burntail,

This issue is now resolved. We have attached Gists for Phi2 and Falcon 1B for your reference. Could you please verify and confirm if we can close the issue?

Thank you!!