unsloth/Phi-3-mini-4k-instruct-v0-bnb-4bit · Silent swapping of Phi-3 mini model

Let me start by thanking you for the amazing Unsloth library: it has been hugely helpful for my work!

I just wanted to mention that I recently found my inference pipelines to be disrupted for no apparent reason, only to discover that the model unsloth/Phi-3-mini-4k-instruct-bnb-4bit I was using as the base model for my adapters was silently moved to this repository. As a consequence, the new version of the model was kinda working but not performing at the same level.

Could you comment on what's different between this version and the updated one? To prevent similar disruptions in the future, it would be very helpful to add disclaimers in the model cards whenever a model is moved or updated!

Hello @danielhanchen @shimmyshimmer

Let me start by thanking you for the amazing Unsloth library: it has been hugely helpful for my work!

I just wanted to mention that I recently found my inference pipelines to be disrupted for no apparent reason, only to discover that the model unsloth/Phi-3-mini-4k-instruct-bnb-4bit I was using as the base model for my adapters was silently moved to this repository. As a consequence, the new version of the model was kinda working but not performing at the same level.

Could you comment on what's different between this version and the updated one? To prevent similar disruptions in the future, it would be very helpful to add disclaimers in the model cards whenever a model is moved or updated!

Hi there thanks for the kind words and apologies for the late reply. Oh ok I see next time we'll just upload a new model with the new version. We just followed what Microsoft did with their Phi-3 models.

This model page you are on is the upgraded version of Phi-3 which Microsoft updated a month ago. The v0 version was the original version of Phi-3.

It's kinda like Llama 3 and Llama 3.1