Models › clef

clef: VRAM needs and speed by GPU

At Q4_K_M, clef needs about 19.1 GB of VRAM. No measured runs yet. Speeds below are estimates. Submit a run.

27.02B parameters · license: apache-2.0 · 262,144-token context · GGUF published Oct 1, 2026 · GGUF files on Hugging Face

Memory needed per quantization

QuantFile sizeNeeds (estimate)Smallest card it fits
Q2_K11.7 GB13.2 GB16 GB cards
Q3_K_S13.6 GB15.1 GB16 GB cards
IQ3_M13.8 GB15.3 GB16 GB cards
Q3_K_M14.5 GB16.0 GB24 GB cards
Q3_K_L15.2 GB16.6 GB24 GB cards
IQ4_XS15.6 GB17.0 GB24 GB cards
Q4_016.2 GB17.7 GB24 GB cards
IQ4_NL16.3 GB17.8 GB24 GB cards
Q4_K_S16.6 GB18.1 GB24 GB cards
Q4_K_M17.7 GB19.1 GB24 GB cards
Q4_117.7 GB19.2 GB24 GB cards
Q4_K_L18.6 GB20.1 GB24 GB cards
Q5_K_S19.6 GB21.1 GB24 GB cards
Q5_K_M20.6 GB22.1 GB24 GB cards
Q6_K23.3 GB24.8 GB32 GB cards
Q6_K_L24.0 GB25.4 GB32 GB cards
Q8_028.8 GB30.3 GB32 GB cards
BF1654.1 GB55.5 GB96 GB cards

Expected speed on popular hardware

HardwareBest quant that fitsExpected tok/s
NVIDIA RTX 3060Does not fit
NVIDIA RTX 4060 TiDoes not fit
NVIDIA RTX 3090Q5_K_M31.8 estimate
NVIDIA RTX 4090Q5_K_M35.2 estimate
NVIDIA RTX 5090Q8_039.9 estimate
AMD Radeon RX 7900 XTXQ5_K_M35.4 estimate
Apple M4 Pro (20-core GPU) (64 GB)Q8_08.0 estimate
Apple M3 Max (40-core GPU) (64 GB)Q8_011.7 estimate
AMD Ryzen AI Max+ 395 (Radeon 8060S)Q8_07.2 estimate

Check your own card

Too big for your card?

Run clef on a rented 24-80 GB GPU by the hour.

Rent a bigger GPU on RunPod

Affiliate link: if you sign up, RigBench may earn a referral credit. It never affects rankings or numbers.