qwen2_audio-mic

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

• Contract manifest
• Transformers usage
• Transformers expectations
• Fixture provenance

Contract

Qwen2-Audio

Runnable usage and the framework expectations directory are linked above.

Slim model

Slim fixture: mic-qwen2_audio-slim

A 20.3M-parameter fixture for the qwen2_audio MIC, covering Qwen/Qwen2-Audio-7B. It is a test artifact, not a usable model, tagged test-artifact to keep it out of candidate-model search.

• Fixture key 098d203cfc53. It covers every checkpoint whose processor files are byte-identical and whose config matches after shrinking.
• Same architecture, smaller config: Qwen2AudioForConditionalGeneration, shrunk by the generic rule in sweep.py (audio encoder_layers 2, text num_hidden_layers 2).
• Processor files copied byte for byte from Qwen/Qwen2-Audio-7B: generation_config.json, merges.txt, preprocessor_config.json, tokenizer.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_features:halve, 0.463), so no training was needed.
• Same tensor names as the real checkpoint. The real checkpoint has every parameter, and so does the fixture.
• Saved in float32, the dtype Qwen/Qwen2-Audio-7B 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 mic-qwen2_audio-slim.