LFM2.5-2.6B

· sglang · 128K context · RTX 3060 Ti · 8 GB

Liquid AI's 2.6B hybrid model: tiny and fast, built for on-device work.

64 tok/sdecode

– tok/sprefill

128Kcontext window

Earlier checkSep 2, 2026

Run it

Follow the steps below to download and run this recipe. Omarchy Local AI lists the recipes it can launch on your hardware. The server then answers on http://localhost:8000/v1.

1 Download the weights

hf download LiquidAI/LFM2.5-2.6B \
  --revision a334ee78cd38458bb71eda24109ac42dcec1309d

2 Start the server

docker run --rm \
  --gpus all \
  -p 8000:30000 \
  --shm-size 16g \
  -e HF_HOME=/root/.cache/huggingface \
  -v ~/.cache/huggingface:/root/.cache/huggingface \
  --entrypoint /opt/sglang/bin/python \
  lmsysorg/sglang:dev-cu12@sha256:c3479164c16d9a8ef33053914c07bfd372007c36c675f629095f2c925c7e7e30 \
  -m sglang.launch_server \
  --model-path LiquidAI/LFM2.5-2.6B \
  --revision a334ee78cd38458bb71eda24109ac42dcec1309d \
  --host 0.0.0.0 \
  --port 30000 \
  --tp 1 \
  --context-length 131072 \
  --mem-fraction-static 0.90 \
  --attention-backend flashinfer \
  --max-running-requests 4 \
  --cuda-graph-max-bs-decode 4 \
  --enable-cache-report \
  --trust-remote-code \
  --reasoning-parser qwen3-thinking \
  --tool-call-parser lfm2

What it passed

✓ loadLoads and serves within an hour

✓ chatAnswers a plain question and stops on its own

– reasoningThinks separately and gets 17 × 23 right

– toolsCalls a tool with the right arguments and uses the result

– contextRecalls a code buried in a prompt that fills 85% of the window

– speedDecodes at 15 tok/s or more

Passed the older check (loads and chats); a full six-check run is pending.

Details

Weights: LiquidAI/LFM2.5-2.6B @ a334ee78cd ›
Image: lmsysorg/sglang:dev-cu12@sha256:c3479164c16d9a8ef33053914c07bfd372007c36c675f629095f2c925c7e7e30
Launch: sglang-lfm2.5-2.6b-bf16-128k-v10 ›
Recipe file: registry/recipes/nvidia/rtx-3060-ti-8gb/lfm2.5-2.6b.sglang.128k.json ›
Model released: Jul 28, 2026
Tested: earlier acceptance