dhanesh-hf/Jarvis-Titan-V15-MoE-Decoupled

Partial fine-tune vs dhanesh-hf/Jarvis-Titan-V14-MoE-Merged. Partial change: 87% of the weights changed (mlp), in 28 of 28 layers.

Tensors changed: Embeddings 0/1, Norms 0/57, MLP 196/196, Attention 0/196, lm_head 0/1

2 derivatives checked

• dhanesh-hf/Jarvis-Titan-M4-Activated: Partial fine-tune
• dhanesh-hf/Jarvis-Titan-M4-MoE-CSA3: Partial fine-tune

On the Hub · Downloads on Model Pulse