BUT-FIT/DiCoW_v3_3
Partial fine-tune (fewer layers) vs openai/whisper-large-v3-turbo. openai/whisper-large-v3-turbo matches part of BUT-FIT/DiCoW_v3_3 (587 of its 861 tensors; the rest of BUT-FIT/DiCoW_v3_3 has no counterpart). On the shared part: Partial change: 79% of the weights changed (attention, embeddings, mlp, norms, other), in 32 of 32 layers.
Tensors changed: Embeddings 1/3, Norms 130/156, Attention 224/280, MLP 128/144, other 4/4