---
tags:
  - mic
  - test-artifact
base_model: facebook/wav2vec2-base-960h
---

# Slim fixture: mic-wav2vec2-slim

A 4.3M-parameter fixture for the [wav2vec2 MIC](https://huggingface.co/hf-internal-testing/wav2vec2-integration), covering [`facebook/wav2vec2-base-960h`](https://huggingface.co/facebook/wav2vec2-base-960h). It is a test artifact, not a usable model, tagged `test-artifact` to keep it out of candidate-model search.

- **Fixture key `447dc0f39f22`.** It covers every checkpoint whose processor files are byte-identical and whose config matches after shrinking.
- **Same architecture, smaller config:** `Wav2Vec2ForCTC`, shrunk by the generic rule in [`sweep.py`](https://github.com/huggingface/model-integration-contracts/blob/main/experiments/sweep.py) (num_hidden_layers 2).
- **Processor files copied byte for byte** from `facebook/wav2vec2-base-960h`: `feature_extractor_config.json`, `preprocessor_config.json`, `special_tokens_map.json`, `tokenizer_config.json`, `vocab.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: `input_values:roll_batch`, 0.603), so no training was needed.
- **Same tensor names as the real checkpoint.** The real checkpoint lacks `wav2vec2.masked_spec_embed`, so the fixture omits it too, and loading initializes it as it does for users.
- **Saved in float32**, the dtype `facebook/wav2vec2-base-960h` loads in by default.
- **Finite in float32, bfloat16, and float16** (forward, loss, gradients).

Built by [`build_fixtures.py`](https://github.com/huggingface/model-integration-contracts/blob/main/example/build_fixtures.py); [`build.json`](build.json) records the pinned sources, gate results, and file hashes. The source repository commits only `README.md` and `build.json`; rebuild the other files with `python example/build_fixtures.py --fixture mic-wav2vec2-slim`.