Working tools / localai.fyi
Memory Workbench
Account for model, cache and runtime memory without making the unknown parts disappear.
All toolsReady. Inputs stay on this device; exports are yours to keep.
Evidence and boundaries
Four retrieved Qwen2.5 configurations are supported. Raw cache arithmetic does not model quantization overhead, device placement, sliding attention, or every runtime allocation.
Documentation retrieved September 15, 2026. These are source facts and stated calculations; visitor measurements are separate.
- llama.cpp server runtime documentation — ggml-org
- Cache strategies · Hugging Face — Hugging Face
- Qwen2.5-0.5B-Instruct configuration — Qwen
- Qwen2.5-1.5B-Instruct configuration — Qwen
- Qwen2.5-3B-Instruct configuration — Qwen
- Qwen2.5-7B-Instruct configuration — Qwen