Add eval.py, README.md, config.json, training_args.json, model.safetensors, USE_POLICY.md
df8b48f verified | license: mit | |
| tags: | |
| - pytorch | |
| - mobilevit | |
| - multitask | |
| - prototype | |
| # Mobilevit experiments: Multitask | |
| ## Overview | |
| Working implementation of **Mobilevit** for **Multitask** using a **large** configuration. The repository focuses on transparent code and repeatable smoke tests; benchmark claims are deliberately omitted. | |
| --- | |
| ## Training details | |
| | Item | Value | | |
| |---|---| | |
| | Framework | PyTorch | | |
| | Precision | fp32 | | |
| | Init scheme | kaiming | | |
| | Hardware | a single consumer GPU is enough | | |
| | Typical run | minutes, not hours | | |
| --- | |
| ## Changelog | |
| 1. 2026-10-05 — initial upload | |
| 2. 2026-10-20 — config and docs sync | |
| --- | |
| ## Acknowledgments | |
| Thanks to the maintainers of the open-source libraries this project leans on. | |
| --- | |
| ## Repository status | |
| One Python module contains the model and a runnable example. | |
| Generated configuration lives in `config.json`; the recipe in `training_args.json`. | |
| The checkpoint is an initialization artifact and should not be read as a trained model. | |
| Scores are deliberately absent. | |
| --- | |
| ## Evaluation guidance | |
| A first evaluation could use a task-specific held-out set. Report the metric across at least three seeds and include a matched-capacity baseline; keep logs and library versions with any published number. | |
| --- | |
| ## Limitations | |
| The initialization checkpoint has not been trained or audited for robustness, fairness, or domain transfer. The implementation should be treated as an experimental starting point. Results from a future trained checkpoint must be documented separately from the defaults shipped here. | |
| --- | |
| ## Default experiment recipe | |
| The generated recipe uses adamw with a polynomial schedule. | |
| These values exist so the script runs end-to-end. A meaningful experiment needs its own tuning budget and at least three seeds. | |
| --- | |
| ## Architecture | |
| | Item | Value | | |
| |---|---| | |
| | Architecture | Mobilevit | | |
| | Scale | large | | |
| | Attention | sparse | | |
| | Fusion | bilinear | | |
| | Activation | swish | | |
| | Normalization | scalenorm | | |
| --- | |
| ## Files | |
| | File | Purpose | | |
| |---|---| | |
| | `eval.py` | primary artifact | | |
| | `README.md` | this documentation | | |
| | `config.json` | architecture configuration | | |
| | `training_args.json` | default experiment settings | | |
| | `model.safetensors` | initialization checkpoint | | |
| | `USE_POLICY.md` | use and redistribution notes | | |
| --- | |
| ## License | |
| Released under **mit**. Review the source-data terms separately when this repository is used with external datasets. | |