The fast language model of the TAPOSA on-device assistant: Qwen3.5-0.8B, fine-tuned by the TAPOSA
project to call the app's 34 tools (calendar, notes, reminders, tasks, alarms, recorded calls).
These files are modified versions of the base model. Apache-2.0, see LICENSE.
| file | what |
|---|---|
taposa-qwen35-0.8b-sft2-qat-Q4_K_M.gguf |
what the app downloads: 4-bit, quantization-aware (529 MB, step success 72.5% on our out-of-distribution benchmark) |
taposa-qwen35-0.8b-sft2-Q4_K_M.gguf |
the same model, plain 4-bit (71.0%) |
Runs in llama.cpp. The generation prompt ends with <think>\n: prepend it before parsing the output.
No account or key is needed to download it.
Qwen/Qwen3.5-0.8B