Harvard-DCML/ADAPT-Olmo3-4.3B-Instruct

Full fine-tune (fewer layers) vs allenai/Olmo-3-7B-Instruct. Harvard-DCML/ADAPT-Olmo3-4.3B-Instruct matches part of allenai/Olmo-3-7B-Instruct (190 of its 355 tensors; the rest of allenai/Olmo-3-7B-Instruct has no counterpart). On the shared part: Fine-tune: 100% of the weights changed, across 17 of 17 layers.

Tensors changed: Embeddings 1/1, MLP 51/51, Norms 69/69, Attention 68/68, lm_head 1/1

On the Hub · Downloads on Model Pulse