jzhang533/PaddleOCR-VL-For-Manga
Partial fine-tune vs PaddlePaddle/PaddleOCR-VL. Partial change: 83% of the weights changed (attention, embeddings, lm_head, mlp, norms), in 27 of 27 layers.
Tensors changed: lm_head 1/1, MLP 155/170, Norms 83/151, Embeddings 3/5, Attention 288/292, other 0/1
2 derivatives checked
- LunarOilRig/PaddleOCR-VL-manga-GGUF: Quantization (fewer layers)
- adambarbato/PaddleOCR-VL-For-Manga-GGUF: Same weights, converted (fewer layers)