Running Liquid AI LFM2.5-2.6B Locally with llama-server

Melvin Vivas · X post · 2026-08-10 · Open on X

Topics: LLMOps, Deployment & Monitoring · Level: intermediate

Summary

This post shares a llama.cpp llama-server command for serving Liquid AI's LFM2.5-2.6B from its Q8_0 GGUF. The server exposes the model on port 8000 with a 128k context. It uses about 6.9 GB of VRAM, so it fits on modest GPUs.

Key points

• Model: LiquidAI/LFM2.5-2.6B-GGUF:Q8_0, using about 6.9 GB VRAM.
• Command: llama-server -hf LiquidAI/LFM2.5-2.6B-GGUF:Q8_0 --alias LiquidAI/LFM2.5-2.6B --port 8000
• --jinja uses the model's chat template; --reasoning-format auto handles the model's reasoning output.
• -ngl all puts all layers on the GPU; -fa on turns on flash attention; -c 128000 sets a 128k context.
• The -hf flag downloads the GGUF straight from Hugging Face.

Resources mentioned

• LFM2.5-2.6B GGUF on Hugging Face (Liquid AI) · tool · huggingface.co · free · open in a browser to verify
  Liquid AI's official GGUF files for the LFM2.5-2.6B small model.
• llama.cpp · repo · github.com · free
  An open-source C/C++ engine for running GGUF models locally. Its llama-server command provides an OpenAI-compatible HTTP server.
  Also in: Running LLMs Locally Without an Expensive Rig (Melvin Vivas on X · notes), llama.cpp / Llama-macOS v0.5.0 release (Melvin Vivas on X · notes), Run llama.cpp GGUF Checkpoints in Hugging Face Transformers (Melvin Vivas on X · notes), llama.cpp v0.4.1 release announcement (Melvin Vivas on X · notes) and 28 more

Try this

• Run the llama-server command to serve LFM2.5-2.6B locally on port 8000.