flowcast-v5-lite v0.2.1
Sub-750MB hybrid LoudInk stack: 99.5% user-demand shipped (126 computer + 74 writing).
Predecessor: nsalerni/flowcast-v4-lite
Hybrid production stack
- Writer longform:
loudink_surgicalโ 100% user-demand writing (shipped) - Writer paraphrase:
paraphrase_posttrainโ 83.3% shipped / 66.7% neural on held-out paraphrase (12 cases) - IR: FunctionGemma IR v5 + transcript-first repairs (github-in-safari, etc.)
- Learned transcript router + polish registry + transcript fast paths
vs v4-lite (user-demand)
| v4-lite | v5-lite hybrid | |
|---|---|---|
| Combined | 65.5% | 99.5% |
| Writing | 9.5% | 100% |
| Computer | 98.4% | 99.2% |
| Paraphrase (held-out) | โ | 83.3% shipped |
| Download | 809 MB | 750 MB |
See inference_config.json and manifest.json for integration fields.
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐ Ask for provider support