nsugianto/tblstructrecog_tuned_tbltransstrucrecog_noncomplex_complex_conlash_b5_1807s_lr1e6_dec1e5_bs4

Full fine-tune vs microsoft/table-transformer-structure-recognition. Fine-tune: 99% of the weights changed, across 6 of 6 layers.

Tensors changed: other 25/110, Norms 63/64, Attention 144/144, MLP 48/48, Embeddings 1/1

On the Hub · Downloads on Model Pulse