Gustav Larsson
Gustav Larsson
I think this is a great feature request that we are definitely thinking about. We have the Open AI CLIP model which features both text and image encoders. However, we...
Thanks for the feature request!
Thank you for the feature request!
Yes, @kory misspoke. We do support v68, and we do support it for some LLMs (Llama 3.2 1B and 3B). You have to specify the precision w4 when exporting (--checkpoint...
How did you compile the assets? If it is through the export script through AI Hub, then they will be compiled with QNN 2.28.2 (you can go to https://app.aihub.qualcomm.com/jobs/ and...
Hi @VaibMittal7, I'm really sorry I dropped the ball on this thread, and we definitely shouldn't have closed it. The most important is that the runtime SDK is a more...
@VaibMittal7 That's really odd. Can you try going to https://qpm.qualcomm.com/ and browsing your way to that package? What does that say?
Error 5000 typically means you compiled with a more recent QAIRT SDK version than you are running it on. Or, you did not target the right device (SoC/hexagon arch). Can...
I don't know when you compiled these assets, which will determine the QAIRT SDK that was used for compilation. If you use 2.38 for your runtime, then it will be...
Hi @VaibMittal7, From your screenshot, it says "libGenie.so version 1.2.0". This is not consistent with QAIRT SDK 2.38. Can you make sure you are not accidentally running an older version?