ArchSpace-Collection/NCP_ArchPreview_dolma3_8.9B_Stage2_v2

Full fine-tune vs ArchSpace-Collection/NCP_ArchPreview_dolma3_8.9B_Stage1. Fine-tune: 100% of the weights changed, across 16 of 16 layers.

Tensors changed: other 351/353, Norms 207/207, MLP 120/120, Attention 160/160, Embeddings 1/1, lm_head 1/1

On the Hub · Downloads on Model Pulse