albert-base-v2-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

ALBERT fill-mask

Runnable usage and the framework expectations directory are linked above.

Slim model

Slim fixture: albert-base-v2-mic

A 11.3M-parameter fixture for the ALBERT MIC, covering albert/albert-base-v2. It is a test artifact, not a usable model, tagged test-artifact to keep it out of candidate-model search.

• Fixture key b903f3096f56. It covers every checkpoint whose processor files are byte-identical and whose config matches after shrinking.
• The checkpoint's own config, not shrunk: AlbertForMaskedLM at 11.3M parameters is at or under the 20M below which a source is kept whole.
• Processor files copied byte for byte from albert/albert-base-v2: spiece.model, tokenizer.json, tokenizer_config.json.
• The checkpoint's own trained weights, unchanged: at 11.3M parameters it is no bigger than a slim fixture. They passed the input-sensitivity gate (weakest: input_ids:roll_batch, 40).
• Same tensor names as the real checkpoint. The real checkpoint has every parameter, and so does the fixture.
• Saved in float32, the dtype albert/albert-base-v2 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 albert-base-v2-mic.