manga-ocr-base-mic

Slim model weights, executable usage and framework expectations in one repository.

• Contract manifest
• Transformers usage
• Transformers expectations
• Fixture provenance
• How the fixture was shrunk

Contract

Manga OCR (VisionEncoderDecoder)

Runnable usage and the framework expectations directory are linked above.

Slim model

Slim fixture: manga-ocr-base-mic

A 0.7M-parameter fixture for the manga-ocr-base MIC, covering kha-white/manga-ocr-base. It is a test artifact, not a usable model, tagged test-artifact to keep it out of candidate-model search.

• Fixture key f52652c64622. It covers every checkpoint whose processor files are byte-identical and whose config matches after shrinking.
• Same architecture, smaller config: VisionEncoderDecoderModel, shrunk by the generic rule in sweep.py ().
• Processor files copied byte for byte from kha-white/manga-ocr-base: preprocessor_config.json, special_tokens_map.json, tokenizer_config.json, vocab.txt.
• Overfit to the contract's training step (50 steps). Random weights failed the input-sensitivity gate: pixel_values:reverse moved the logits by only 1.3e-05 (threshold 0.001).
• Same tensor names as the real checkpoint. The real checkpoint has every parameter, and so does the fixture.
• Saved in float32, the dtype kha-white/manga-ocr-base loads in by default.
• Finite in float32, bfloat16, and float16 (forward, loss, gradients).

Built by build_fixtures.py; build.json records the pinned sources, gate results, and file hashes. This repository includes the slim weights, usage and framework expectations; rebuild the other files with python example/build_fixtures.py --fixture manga-ocr-base-mic.