Fit receipt — a fixed address for one verdict

Qwen 3.8 27B on Mac 32GB

WON'T FIT ✗

Conditions: 8-bit · 8K context · KV F16

weights	KV cache	linear state	overhead	reserve	total	short by
25.9 GB	0.5 GB	0.1 GB	3.4 GB	8.5 GB	38.4 / 32 GB	6.4 GB

→ Drop the weights to 4-bit and it fits (KV unchanged).

Replay: npx fitllm "Qwen 3.8 27B" --mac 32 --quant 8 --ctx 8192 --kv 16 · JSON · interactive

live fit badge — embed: ![fits](https://img.shields.io/endpoint?url=https%3A%2F%2Ffitllm.run%2Fapi%2Fbadge%3Fmodel%3DQwen%2B3.8%2B27B%26ram%3D32%26quant%3D8%26ctx%3D8192%26kv%3D16)

Computed by the open fitllm-engine (MIT) from official config.json values — estimates; runtime varies. Ran it for real? Challenge this prediction — measured reports calibrate the engine for everyone.

engine 2.21.4 · revision 2dca33db8c822bb2348949b5c54797d86a869202 · release manifest