Haon-Chen/e5-omni-7B
Partial fine-tune vs Qwen/Qwen2.5-Omni-7B. Partial change: 33% of the weights changed (attention, mlp), in 28 of 32 layers.
Tensors changed: other 0/7, MLP 12/336, Norms 0/203, Attention 64/556, Embeddings 0/1, lm_head 0/1
Partial fine-tune vs Qwen/Qwen2.5-Omni-7B. Partial change: 33% of the weights changed (attention, mlp), in 28 of 32 layers.
Tensors changed: other 0/7, MLP 12/336, Norms 0/203, Attention 64/556, Embeddings 0/1, lm_head 0/1