Fit receipt — a fixed address for one verdict

GLM-4.7-Flash on RTX 3090

FITS ✓

Conditions: Q4_K_M · 8K context · KV F16 · Windows + display reserve

weights	KV cache	overhead	reserve	total	free
17.8 GB	0.4 GB	2.4 GB	2.0 GB	22.6 / 24 GB	1.4 GB

Context at this quant: recommended max ~10K tokens (keeps 5% headroom) · hard limit ~24K tokens (beyond it, it won't fit)

Active structural premises

mla-compressed-latent-cache — KV memory assumes a compressed-latent MLA artifact or mode; legacy non-MLA GGUF or an explicitly uncompressed mode invalidates this estimate.

Replay: npx fitllm "GLM-4.7-Flash" --gpu "RTX 3090" --quant Q4_K_M --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%3DGLM-4.7-Flash%26gpu%3DRTX%2B3090%26quant%3DQ4_K_M%26env%3Dwindows-display%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