Na0s/Mixtral-8x7B-v0.1-instruct-l2-norm-post-Gates-LoRA-SFT-pruned-1-expert
LoRA-style vs mistralai/Mixtral-8x7B-Instruct-v0.1. Looks like a merged LoRA (or a fine-tune with frozen norms): only the weight matrices of w1, w2, w3 changed, in 28 of 32 layers; norms, biases and embeddings are untouched.
Tensors changed: Embeddings 0/1, MLP 330/672, Norms 0/65, Attention 0/128, lm_head 0/1