rAIfle/SorcererLM-8x22b-bf16

Full fine-tune vs alpindale/WizardLM-2-8x22B. Fine-tune: 100% of the weights changed, across 56 of 56 layers.

Tensors changed: Embeddings 0/1, MLP 1344/1344, other 56/56, Attention 224/224, Norms 0/113, lm_head 0/1

1 derivatives checked

• Quant-Cartel/SorcererLM-8x22b-iMat-GGUF: Quantization

1 other repos hold byte-for-byte identical weight files.

On the Hub · Downloads on Model Pulse