deepvk/llava-saiga-8b

Full fine-tune (fewer layers) vs IlyaGusev/saiga_llama3_8b. IlyaGusev/saiga_llama3_8b matches part of deepvk/llava-saiga-8b (289 of its 686 tensors; the rest of deepvk/llava-saiga-8b has no counterpart). On the shared part: Fine-tune: 100% of the weights changed, across 32 of 32 layers.

Tensors changed: Norms 65/65, MLP 96/96, Attention 128/128

On the Hub · Downloads on Model Pulse