lewtun/tailsft-repro-smoke-tail50

Partial fine-tune vs allenai/OLMo-2-0425-1B. Partial change: 72% of the weights changed (attention, mlp, norms), in 16 of 16 layers.

Tensors changed: Embeddings 0/1, MLP 48/48, Norms 38/65, Attention 64/64, lm_head 0/1

On the Hub · Downloads on Model Pulse