Quantization made by Richard Erkhov.
pruned_mistral - bnb 8bits
Original model description:
library_name: transformers
datasets:
- HuggingFaceH4/ultrachat_200k
- HuggingFaceH4/ultrafeedback_binarized
base_model: mistralai/Mistral-7B-v0.1
license: apache-2.0
A Pruned Mistral model
This model is a pruned Mistral model re-aligned using the Zephyr Recipe
details
- This model has 2 stages training: SFT and DPO
- The initial model consist on selecting some layers of the mistral model to make a smaller model
- the code can be found here: github.com/tcapelle/shear
W&B workspace
