Models › LFM2.5-8B-A1B

LFM2.5-8B-A1B: VRAM needs and speed by GPU

At UD-Q4_K_M, LFM2.5-8B-A1B needs about 6.3 GB of VRAM. No measured runs yet. Speeds below are estimates. Submit a run.

8.47B parameters (mixture-of-experts, ~1.1B active) · license: other · 128,000-token context · GGUF published May 29, 2026 · GGUF files on Hugging Face

Memory needed per quantization

QuantFile sizeNeeds (estimate)Smallest card it fits
UD-IQ1_M2.6 GB3.6 GB8 GB cards
UD-IQ2_XXS2.7 GB3.7 GB8 GB cards
UD-IQ2_M2.8 GB3.8 GB8 GB cards
UD-Q2_K_XL2.9 GB3.9 GB8 GB cards
UD-IQ3_XXS3.3 GB4.3 GB8 GB cards
UD-IQ3_S3.6 GB4.6 GB8 GB cards
UD-Q3_K_M3.9 GB4.9 GB8 GB cards
UD-Q3_K_XL4.0 GB5.0 GB8 GB cards
UD-IQ4_XS4.3 GB5.3 GB8 GB cards
UD-IQ4_NL4.4 GB5.4 GB8 GB cards
UD-Q4_K_S5.0 GB6.0 GB8 GB cards
UD-Q4_K_M5.3 GB6.3 GB8 GB cards
MXFP4_MOE5.3 GB6.4 GB8 GB cards
UD-Q4_K_XL5.4 GB6.4 GB8 GB cards
UD-Q5_K_S6.0 GB7.0 GB8 GB cards
UD-Q5_K_M6.4 GB7.4 GB8 GB cards
UD-Q5_K_XL6.4 GB7.4 GB8 GB cards
UD-Q6_K7.1 GB8.1 GB12 GB cards
UD-Q6_K_XL7.7 GB8.8 GB12 GB cards
Q8_09.0 GB10.0 GB12 GB cards
UD-Q8_K_XL9.3 GB10.4 GB12 GB cards
BF1616.9 GB18.0 GB24 GB cards

Expected speed on popular hardware

HardwareBest quant that fitsExpected tok/s
NVIDIA RTX 3060Q8_0137 estimate
NVIDIA RTX 4060 TiUD-Q5_K_M154 estimate
NVIDIA RTX 3090Q8_0291 estimate
NVIDIA RTX 4090Q8_0323 estimate
NVIDIA RTX 5090Q8_0510 estimate
AMD Radeon RX 7900 XTXQ8_0324 estimate
Apple M4 Pro (20-core GPU) (64 GB)Q8_0103 estimate
Apple M3 Max (40-core GPU) (64 GB)Q8_0150 estimate
AMD Ryzen AI Max+ 395 (Radeon 8060S)Q8_092.3 estimate

Check your own card

Too big for your card?

Run LFM2.5-8B-A1B 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.