plaintext
722 chars · 12 lines
[2] ESTIMATED PERFORMANCE ON THIS MACHINE (22K–35K Active Agent Context)
• Slot 0 Disk Snapshot Restore (22.2K tok) : ~10.3 ms
• Warm Interactive TTFT (40-tool prefix) : ~0.51 s
• Cold Prompt Prefill Throughput : ~183 tok/s
• Decode — Base Physical Floor (no spec) : ~18.5 tok/s
• Decode — Technical Prose + New Code : ~21.7 tok/s (RouteCrack ~78% acceptance)
• Decode — Agent <tool_call> / JSON / XML : ~26.6 tok/s
• Decode — Refactoring / Boilerplate Burst : ~45.1 tok/s
• Max Concurrent 256K Agent Workers : 0 instance(s)
curl -sL https://raw.githubusercontent.com/synchopate/llama-modchip/main/scripts/check_cpu_readiness.py | python3