LFM2.5-8B-A1B

Liquid AI · liquid/lfm2-5-8b-a1b

ProviderLiquid AI
Familylfm2
Typellm-chat
Statusactive
Released2026-05-28
Updated2026-08-24
Parameters8.5B
Open weightsyes

Lineage

RelationshipModel
Is a finetune ofLiquidAI/LFM2.5-8B-A1B-Base

What it fits on

Ordered by predicted speed. Bandwidth sets decode rate; memory decides whether it runs at all.

Every figure here is computed, not measured. Fit is weights at each quantisation against device memory, with a 25% allowance for the KV cache, activations and the OS. Decode rate is the memory-bandwidth roofline at 70% efficiency. Nobody has run this model on these devices. This is a mixture-of-experts model and no card carries active-parameter counts, so these predictions use total parameters and understate the real speed.
DeviceMemoryBandwidthBest qualityWeightstok/sFastest
Cerebras WSE-344.0 GB21000000.0 GB/sbf1616.94 GB~867988.2~3471952.9 q4
NVIDIA GB200 Grace Blackwell Superchip372.0 GB16000.0 GB/sbf1616.94 GB~661.3~2645.3 q4
AMD Instinct MI355X288.0 GB8000.0 GB/sbf1616.94 GB~330.7~1322.6 q4
NVIDIA B200180.0 GB7700.0 GB/sbf1616.94 GB~318.3~1273.0 q4
Google TPU7x (Ironwood)192.0 GB7380.0 GB/sbf1616.94 GB~305.0~1220.1 q4
AMD Instinct MI325X256.0 GB6000.0 GB/sbf1616.94 GB~248.0~992.0 q4
AMD Instinct MI300X192.0 GB5300.0 GB/sbf1616.94 GB~219.1~876.3 q4
NVIDIA H200 SXM141.0 GB4800.0 GB/sbf1616.94 GB~198.4~793.6 q4
NVIDIA H100 NVL94.0 GB3900.0 GB/sbf1616.94 GB~161.2~644.8 q4
NVIDIA H100 SXM80.0 GB3350.0 GB/sbf1616.94 GB~138.5~553.9 q4
AMD Instinct MI250X128.0 GB3200.0 GB/sbf1616.94 GB~132.3~529.1 q4
Google TPU v5p95.0 GB2765.0 GB/sbf1616.94 GB~114.3~457.1 q4
NVIDIA A100 80GB SXM80.0 GB2039.0 GB/sbf1616.94 GB~84.3~337.1 q4
NVIDIA H100 PCIe80.0 GB2000.0 GB/sbf1616.94 GB~82.7~330.7 q4
NVIDIA GeForce RTX 509032.0 GB1792.0 GB/sbf1616.94 GB~74.1~296.3 q4
Google TPU v6e (Trillium)32.0 GB1638.0 GB/sbf1616.94 GB~67.7~270.8 q4
AMD Instinct MI21064.0 GB1600.0 GB/sbf1616.94 GB~66.1~264.5 q4
NVIDIA A100 40GB SXM40.0 GB1555.0 GB/sbf1616.94 GB~64.3~257.1 q4
Google TPU v432.0 GB1200.0 GB/sbf1616.94 GB~49.6~198.4 q4
NVIDIA GeForce RTX 409024.0 GB1008.0 GB/sbf1616.94 GB~41.7~166.7 q4
AMD Radeon RX 7900 XTX24.0 GB960.0 GB/sbf1616.94 GB~39.7~158.7 q4
NVIDIA GeForce RTX 508016.0 GB960.0 GB/sfp88.47 GB~79.4~158.7 q4
NVIDIA GeForce RTX 309024.0 GB936.0 GB/sbf1616.94 GB~38.7~154.7 q4
NVIDIA L40S48.0 GB864.0 GB/sbf1616.94 GB~35.7~142.8 q4
Apple M3 Ultra512.0 GB819.0 GB/sbf1616.94 GB~33.9~135.4 q4
AMD Radeon RX 7900 XT20.0 GB800.0 GB/sfp88.47 GB~66.1~132.3 q4
Apple M2 Ultra192.0 GB800.0 GB/sbf1616.94 GB~33.1~132.3 q4
Google TPU v5e16.0 GB800.0 GB/sfp88.47 GB~66.1~132.3 q4
NVIDIA GeForce RTX 4080 SUPER16.0 GB736.0 GB/sfp88.47 GB~60.8~121.7 q4
NVIDIA GeForce RTX 4070 Ti SUPER16.0 GB672.0 GB/sfp88.47 GB~55.6~111.1 q4
AMD Radeon RX 9070 XT16.0 GB640.0 GB/sfp88.47 GB~52.9~105.8 q4
Apple M4 Max128.0 GB546.0 GB/sbf1616.94 GB~22.6~90.3 q4
Apple M1 Max64.0 GB400.0 GB/sbf1616.94 GB~16.5~66.1 q4
Apple M2 Max96.0 GB400.0 GB/sbf1616.94 GB~16.5~66.1 q4
Apple M3 Max128.0 GB400.0 GB/sbf1616.94 GB~16.5~66.1 q4
NVIDIA GeForce RTX 3060 12GB12.0 GB360.0 GB/sfp88.47 GB~29.8~59.5 q4
NVIDIA L424.0 GB300.0 GB/sbf1616.94 GB~12.4~49.6 q4
NVIDIA GeForce RTX 4060 Ti 16GB16.0 GB288.0 GB/sfp88.47 GB~23.8~47.6 q4
Apple M4 Pro64.0 GB273.0 GB/sbf1616.94 GB~11.3~45.1 q4
Apple M432.0 GB120.0 GB/sbf1616.94 GB~5.0~19.8 q4

Where it runs

Taken from the card's availability section.

PlatformId on that platform
Hugging FaceLiquidAI/LFM2.5-8B-A1B

Verified benchmark evidence

Each row was checked against its source by a reviewer, and carries the model identifier as actually evaluated — which is not always the same as this card's.
BenchmarkEvaluated asScoreEvidence dateSource kind
aa_lcrLFM2.5-8B-A1B0.0%2026-09-10 evaluatedindependent evaluatorsource
critptLFM2.5-8B-A1B0.0%2026-09-10 evaluatedindependent evaluatorsource
gpqa_diamondLFM2.5-8B-A1B51.31%2026-09-10 evaluatedindependent evaluatorsource

Reported benchmark scores

This card reports no benchmark scores yet.

BenchmarkCatalogue standingScore

Data

This card as JSON · See it in the graph · Edit on GitHub