SupraLabs/Supra1.5-50M-Instruct-exp

Full fine-tune vs SupraLabs/Supra1.5-50M-Base-exp. Fine-tune: 100% of the weights changed, across 12 of 12 layers.

Tensors changed: Embeddings 1/1, Norms 1/25, MLP 36/36, Attention 48/48

1 other repos hold byte-for-byte identical weight files.

On the Hub · Downloads on Model Pulse