41 t/s generation and 570+ t/s ingest on a 16gb laptop box.
flash attention + full gpu layer offload on llama.cpp, wired directly to @interaction mcp.
cooking something devious
Happy 250th Birthday, America! πΊπΈ
To celebrate, @Interaction is making Poke Human available for everyone this weekend.
To try:
1. Text Poke a task AI can't do yet
2. Poke dispatches a Human subagent
3. Your task gets accomplished β