Models › Llama-3.2-3B-Instruct-uncensored

Llama-3.2-3B-Instruct-uncensored: VRAM needs and speed by GPU

At Q4_K_M, Llama-3.2-3B-Instruct-uncensored needs about 3.1 GB of VRAM. No measured runs yet. Speeds below are estimates. Submit a run.

3.61B parameters · 131,072-token context · GGUF published Oct 26, 2024 · GGUF files on Hugging Face

Memory needed per quantization

QuantFile sizeNeeds (estimate)Smallest card it fits
Q2_K1.5 GB2.4 GB8 GB cards
IQ3_XS1.6 GB2.5 GB8 GB cards
Q3_K_S1.7 GB2.6 GB8 GB cards
IQ3_M1.8 GB2.7 GB8 GB cards
Q3_K_M1.9 GB2.7 GB8 GB cards
Q2_K_L1.9 GB2.8 GB8 GB cards
Q3_K_L2.0 GB2.9 GB8 GB cards
IQ4_XS2.0 GB2.9 GB8 GB cards
Q4_02.1 GB3.0 GB8 GB cards
Q4_K_S2.1 GB3.0 GB8 GB cards
Q4_K_M2.2 GB3.1 GB8 GB cards
Q3_K_XL2.3 GB3.2 GB8 GB cards
Q4_K_L2.5 GB3.4 GB8 GB cards
Q5_K_S2.5 GB3.4 GB8 GB cards
Q5_K_M2.6 GB3.5 GB8 GB cards
Q5_K_L2.8 GB3.7 GB8 GB cards
Q6_K3.0 GB3.9 GB8 GB cards
Q6_K_L3.2 GB4.0 GB8 GB cards
Q8_03.8 GB4.7 GB8 GB cards
F167.2 GB8.1 GB12 GB cards

Expected speed on popular hardware

HardwareBest quant that fitsExpected tok/s
NVIDIA RTX 3060Q8_080.2 estimate
NVIDIA RTX 4060 TiQ8_063.7 estimate
NVIDIA RTX 3090Q8_0171 estimate
NVIDIA RTX 4090Q8_0189 estimate
NVIDIA RTX 5090Q8_0299 estimate
AMD Radeon RX 7900 XTXQ8_0190 estimate
Apple M4 Pro (20-core GPU) (64 GB)Q8_060.3 estimate
Apple M3 Max (40-core GPU) (64 GB)Q8_088.0 estimate
AMD Ryzen AI Max+ 395 (Radeon 8060S)Q8_054.1 estimate

Check your own card

Too big for your card?

Run Llama-3.2-3B-Instruct-uncensored 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.