| Gemma 4 on Cactus |
Henry Ndubuaku |
Native multimodal voice, vision, and audio on-device with hybrid cloud handoff |
| Hybrid Transcription |
Roman Shemet |
Sub-150ms transcription with cloud-level accuracy using on-device/cloud hybrid inference |
| On-Device Coding Agents |
Noah Cylich & Henry Ndubuaku |
Running LFM2-24B MoE locally on Mac for coding use cases |
| Ridiculously Fast Transcription |
Satyajit Kumar & Henry Ndubuaku |
6M tok/sec decode speed with Parakeet CTC 1.1B |
| LFM-2.5-350m on Cactus |
Henry Ndubuaku |
140 tok/sec single-core INT8 inference across seven devices, from Vision Pro to Raspberry Pi 5 |