PaddlePaddle/HPD-Parsing

Full fine-tune vs OpenGVLab/InternVL3_5-1B. Fine-tune: 100% of the weights changed, across 28 of 28 layers.

Tensors changed: Norms 209/209, MLP 186/186, Attention 208/208, Embeddings 4/4, other 48/48

1 other repos hold byte-for-byte identical weight files.

On the Hub · Downloads on Model Pulse