mask2former-swin-tiny-coco-instance-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

Mask2Former instance segmentation

Runnable usage and the framework expectations directory are linked above.

Slim model

Slim fixture: mask2former-swin-tiny-coco-instance-mic

A 8.0M-parameter fixture for the Mask2Former MIC, covering facebook/mask2former-swin-tiny-coco-instance. It is a test artifact, not a usable model, tagged test-artifact to keep it out of candidate-model search.

• Fixture key 6cdde2aa8848. It covers every checkpoint whose processor files are byte-identical and whose config matches after shrinking.
• Same architecture, smaller config: Mask2FormerForUniversalSegmentation, shrunk by the generic rule in sweep.py (encoder_layers 2, decoder_layers 2, backbone num_layers 2).
• Processor files copied byte for byte from facebook/mask2former-swin-tiny-coco-instance: preprocessor_config.json.
• Random weights (seed 17). They passed the input-sensitivity gate: every perturbation of the contract's inputs moved the logits by at least 0.001 (weakest: pixel_values:halve, 0.00101), so no training was needed.
• Same tensor names as the real checkpoint. The real checkpoint lacks model.pixel_level_module.encoder.swin.layernorm.bias, model.pixel_level_module.encoder.swin.layernorm.weight, so the fixture omits it too, and loading initializes it as it does for users.
• Saved in float32, the dtype facebook/mask2former-swin-tiny-coco-instance 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 mask2former-swin-tiny-coco-instance-mic.