Dismatch of this model and the model from nvidia/llama-3_2-nv-embedqa-1b-v2

#2
by mzl - opened

I just test the embedding of this model and compare the result from nvidia/llama-3_2-nv-embedqa-1b-v2. The vectors are different. What's the reason?

@mzl Please can you provide more details on the reference you're comparing to? Are you referring to the hosted version running here https://build.nvidia.com/nvidia/llama-3_2-nv-embedqa-1b-v2 or something else?

I suspect this may have been related to the version of transformers used to load the model . We have made a change in #16 to improve support for more transformers versioins which should mitigate this issue. Closing this issue now.

Feel free to open a new one with more details if there is still an issue with latest version.

nvidia-oliver-holworthy changed discussion status to closed

Sign up or log in to comment