Load failed: 34
I copy pasted the .pte file to /data/local/tmp/llama/ and also put the tokenizer from llama3 repo, tried multiple builds, how to make it work?
These PTE files are compiled to be used with Layla - the offline assistant: https://www.layla-network.ai/
You can download the Executorch app in Layla and load the model using the RMI (Raw Model Instructions app). Please note this only works on very very high end phones, such as the S23 S24 Ultra.
Hello
@l3utterfly
I got your executorch LLAMA 3 model to work yesterday perfectly on latest version of Layla on a Samsung Fold 3, I tried to run it again today and its giving this error "Load failed: 34"
I've enabled logs but whenever I try to check logs it says log is Empty...
@yehiaserag Try the normal llama.cpp quant and the latest version of Layla: https://huggingface.co/mmnga/lightblue-suzume-llama-3-8B-multilingual-gguf/tree/main (try the Q4_K_M quant). Executorch is a little outdated in Layla currently because llama.cpp is currently better in all aspects (speed, QoL, etc.)
I will update Executorch to the latest version soon.
Thanks a lot for the swift response