Text Generation
Transformers
Safetensors
zamba2
Inference Endpoints

Was the new PHI-2.7B used in the Comparison?

#1
by S4sch - opened

I'm wondering, whether in the comparison the new or the old Phi-2.7b was used, considering that it was updated quite silently without a new version number.

We used the phi2-2.7B from Microsoft's HF repo: https://huggingface.co/microsoft/phi-2/tree/main

I see that those model weights were updated ~7 months ago, so I'm certain we used their latest version for evals.

If your question is about comparing evals to phi3-3.8B at https://huggingface.co/microsoft/Phi-3-mini-128k-instruct, we didn't compare because our model is base and theirs is instruct-tuned. If Microsoft releases the base model for phi3-3.8B we'll be happy to compare. Otherwise that comparison will have to wait until we release our instruct-tuned zamba2.

Sign up or log in to comment