Spaces:
Running
Running
| title: Heed - 80% of Sonnet on a phone | |
| emoji: 🎙️ | |
| colorFrom: green | |
| colorTo: gray | |
| sdk: static | |
| pinned: true | |
| license: apache-2.0 | |
| short_description: How a 0.6B on-device model reached 80% of Claude Sonnet | |
| models: | |
| - mohitkrconnect/heed-tasks-qwen3-0.6b | |
| tags: | |
| - on-device | |
| - task-extraction | |
| - small-language-model | |
| - research | |
| # Heed study | |
| How the Heed task model was built and evaluated: the sealed test, the LLM judge and how it was checked against a human, the metric, the data and the training. | |