LFM2.5 (2.6B, Q4)
Liquid AI's LFM2.5 at 2.6B parameters. A hybrid convolution-attention design built for on-device speed — light on memory, quick even on modest hardware, with integrated reasoning, native tool calling, and a 128K context window.
At a glance
- Parameters
- 2.6B
- Quantization
- Q4_K_M
- Context window
- 128K tokens
- Approx. size
- 1.7 GB
- Engines
- llama.cpp, MLX
- License
- LFM Open License v1.0 (custom-restricted)
- Version
- v1.0.1 , released 2026-08-15
- Family
- other
Tuning
Defaults a gezel client applies out of the box, from real evaluation runs against this exact quantization.
- sampling.temperature
0.1- sampling.topK
50- sampling.repetitionPenalty
1.1- reasoning.thinkingBudget
2048- reasoning.enableThinking
true- profiles
- thinking-general, thinking-coding, thinking-precise, instruct, creative
Model behaviors
Client-side behavior modules the catalog enables for this model (reasoning-tag handling, fabrication detection, prompt shaping).
- reasoning.strip-think-tags
- prompt.private-reasoning-guidance
- prompt.tool-cookbook-full
- fabrication.detect-past-tense-no-tools
- mcp.compact-tool-schemas
- mcp.relax-required-fields
- mcp.default-missing-fields
- turn.continuation-budget
- turn.ollama-num-predict-bumped
- turn.preamble-folding
- turn.ramble-detection
- prompt.prefer-writefile-edits
- prompt.derive-by-execution
- prompt.native-tool-call-format
Sources
- llama.cpp (GGUF)
- LiquidAI/LFM2.5-2.6B-GGUF ·
LFM2.5-2.6B-Q4_K_M.gguf· Q4_K_M - MLX
- mlx-community/LFM2.5-2.6B-4bit
- Upstream
- https://huggingface.co/LiquidAI/LFM2.5-2.6B
- Manifest
- View on GitHub
Tags
- liquid-ai
- small
- tools