switch-base-8-mic / README.md
tarekziade's picture
tarekziade HF Staff
Publish MIC usage, slim fixture and reviewed Transformers expectations
0e526fe verified
|
Raw History Blame Contribute Delete
2.89 kB
---
tags:
- mic
- test-artifact
base_model: google/switch-base-8
---
# switch-base-8-mic
Slim model weights, executable usage and framework expectations in one repository.
- [Contract manifest](integration.json)
- [Transformers usage](integrations/transformers/usage.md)
- [Transformers expectations](integrations/transformers/expectations)
- [Fixture provenance](build.json)
- [How the fixture was shrunk](https://huggingface-mic-mic-coverage.static.hf.space/shrink/switch-base-8-mic.html)
## Contract
## Slim model
### switch-base-8-mic
Slim model weights, executable usage and framework expectations in one repository.
- [Contract manifest](integration.json)
- [Transformers usage](integrations/transformers/usage.md)
- [Transformers expectations](integrations/transformers/expectations)
- [Fixture provenance](build.json)
- [How the fixture was shrunk](https://huggingface-mic-mic-coverage.static.hf.space/shrink/switch-base-8-mic.html)
## Contract
### SwitchTransformers text-to-text generation
Runnable usage and the framework expectations directory are linked above.
## Slim model
### Slim fixture: switch-base-8-mic
A 2.3M-parameter fixture for the [SwitchTransformers MIC](https://huggingface.co/huggingface-mic/switch-base-8-mic), covering [`google/switch-base-8`](https://huggingface.co/google/switch-base-8). It is a test artifact, not a usable model, tagged `test-artifact` to keep it out of candidate-model search.
- **Fixture key `c3577706d546`.** It covers every checkpoint whose processor files are byte-identical and whose config matches after shrinking.
- **Same architecture, smaller config:** `SwitchTransformersForConditionalGeneration`, shrunk by the generic rule in [`sweep.py`](https://github.com/huggingface/transformers-mic/blob/main/experiments/sweep.py) (num_layers 2, num_decoder_layers 2).
- **Processor files copied byte for byte** from `google/switch-base-8`: `generation_config.json`, `special_tokens_map.json`, `spiece.model`, `tokenizer.json`, `tokenizer_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: `attention_mask:all_ones`, 4.36), 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 bfloat16**, the dtype `google/switch-base-8` loads in by default.
- **Finite in float32, bfloat16, and float16** (forward, loss, gradients).
Built by [`build_fixtures.py`](https://github.com/huggingface/transformers-mic/blob/main/example/build_fixtures.py); [`build.json`](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 switch-base-8-mic`.