dphn/dolphin-2.9-mixtral-8x22b
LoRA-style vs mistral-community/Mixtral-8x22B-v0.1. Looks like a merged LoRA (or a fine-tune with frozen norms): only the weight matrices of k_proj, o_proj, q_proj, v_proj, w1, w2, w3 changed, in 33 of 56 layers; norms, biases and embeddings are untouched.
Tensors changed: MLP 336/1344, other 0/56, Attention 56/224, Norms 0/113
1 derivatives checked
- mradermacher/dolphin-2.9-mixtral-8x22b-i1-GGUF: Quantization