models · liquidai

LFM2.5-2.6B

text-generation model harvested from Hugging Face (trending). Catalogued at Quarantine — unverified, pending attestation.

Compare Verify

What this registry has checked

Hardened. We downloaded these weights and hashed them ourselves. The hash below is ours, not a publisher’s claim. It does not assert that the model is safe, accurate, or fit for any purpose. It says the bytes are the bytes.

Method	self-download-sha256

Re-check this yourself → Source → Full record →

What is recorded

Publisher	liquidai
Origin	United States
Licence	other
Parameters	2,697,198,592 (2.7B)
Context window	65,536 tokens
Modality	text->text
Architecture	lfm2
Layers	30
Hidden size	2048
Attention	GQA (32 heads, 8 KV)
Native precision	BF16
Pulls recorded here	47,393
Downloads reported upstream	98,713

Reported scores

GPQA Diamond	55.8
Humanity's Last Exam	6.2

Reported by a third party and recorded here, not re-run by us. Source: artificialanalysis.ai/api/v2.

Measured on real hardware

None of these runs are ours. Each figure is the reporter’s, linked to their own evidence, under their own trust label. The number shown is single-stream decode — one user, concurrency 1 — because aggregate throughput across many users is a different quantity and reads as far faster than anyone will see.

Reported to local-ai-registry

Hardware	Engine	Build measured	Decode, 1 stream	Label	Reported by
GeForce RTX 3080 Ti	ollama	Q8_0 Q4_K_M	277 tok/s @ 32,768	candidate	0xsero
H200 141GB	ollama	Q8_0 Q4_K_M	206 tok/s @ 32,768	candidate	0xsero
GeForce RTX 3070	ollama	Q8_0 Q4_K_M	186 tok/s @ 16,384	candidate	0xsero
GeForce RTX 3060 Ti	ollama	Q8_0 Q4_K_M	172 tok/s @ 16,384	candidate	0xsero
Tesla V100 16GB	ollama	Q8_0 Q4_K_M	156 tok/s @ 32,768	candidate	0xsero
Intel Arc Pro B70	llama.cpp	Q8_0	132 tok/s @ 8,192	candidate	localmaxxing
GeForce RTX 4060 Ti	ollama	Q8_0 Q4_K_M	131 tok/s @ 32,768	candidate	0xsero
RTX 6000 Ada Generation	ollama	Q8_0 Q4_K_M	125 tok/s @ 32,768	candidate	0xsero
NVIDIA GeForce RTX 4060 Laptop GPU	llama.cpp	GGUF QAD-Q4_0	120 tok/s @ 4,076	validated	jaytau
RTX A5000 24GB	ollama	Q8_0 Q4_K_M	91.2 tok/s @ 32,768	candidate	0xsero

From local-ai-registry (MIT), commit 124959d: 57 runs across 41 machines, 18 validated — model revision and runtime pinned, launch accepted — and 39 candidate, their label for useful evidence without a reproducible-launch promise. 9 more not listed. 19 report figures that contradict each other — prefill below decode, which parallel prefill essentially never is — and are left to the source rather than ranked here.

Elsewhere in the registry

All models → Browse with filters → What will run on my hardware? →

Registry record as of 2026-09-01. Identity, licence and parameter count are properties of a release and do not move; tier can, and the verify link above re-checks it against the live registry rather than this page.