UCSC-VLAA/MedVLThinker-32B-RL_m23k
Full fine-tune vs Qwen/Qwen2.5-VL-32B-Instruct. Fine-tune: 94% of the weights changed, across 64 of 64 layers.
Tensors changed: Attention 448/576, MLP 192/388, Norms 127/194, Embeddings 0/2, lm_head 1/1
Full fine-tune vs Qwen/Qwen2.5-VL-32B-Instruct. Fine-tune: 94% of the weights changed, across 64 of 64 layers.
Tensors changed: Attention 448/576, MLP 192/388, Norms 127/194, Embeddings 0/2, lm_head 1/1