manga-ocr-finetuned (ONNX)

ONNX export of JustANormalTinkerer/manga-ocr-finetuned โ€” a finetuned manga-OCR with a distilled 192-dim ViT encoder (the upstream ships PyTorch only).

  • Weights: encoder_model.onnx (22 MB, 192-dim encoder) + decoder_model.onnx (118 MB, non-merged), exported with optimum. I/O: encoder pixel_values โ†’ last_hidden_state; decoder input_ids + encoder_hidden_states โ†’ logits.
  • Tokenizer: the manga-ocr 6144-token WordPiece tokenizer.json, verified vocab-identical to this model's lineage.

Repackaged for single-repo, self-consistent ONNX inference.

Downloads last month
7
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for NorwayFish/manga-ocr-finetuned

Quantized
(1)
this model