logic65/whittle-next-moe-test
Full fine-tune vs logic65/Qwen3.8-Whittle-tri-14.7B. logic65/Qwen3.8-Whittle-tri-14.7B is a GGUF (Q8) of different weights. Fine-tune: 100% of the weights changed, across 30 of 32 layers.
Tensors changed: lm_head 1/1, Norms 56/81, Embeddings 1/1, Attention 32/32