General Questions
What are LFM models?
What are LFM models?
LFM (Liquid Foundation Models) are Liquid AIβs efficient models for local and hosted deployment. LFM3 is the latest generation, with text, vision, audio, omni, and decision models. The library also includes smaller and specialized LFM2.5 and LFM2 models, including embedding and encoder models. Open-weight models range from 230M to 24B parameters.
What context length do LFM models support?
What context length do LFM models support?
LFM3-1.7B supports 256K tokens, and LFM3-VL-2B supports 128K. Other limits vary by model: many LFM text, vision, and audio models support 32K, LFM2.5-8B-A1B supports 128K, and LFM2.5-24B-A2B supports 131K. Encoder-based models have shorter limits (8K for the LFM2.5 encoders, 16K for d1-omni-600M). Each model page lists its exact limit; hosted API limits can differ from the open-weight checkpoints.
Which inference frameworks are supported?
Which inference frameworks are supported?
LFM models are compatible with the frameworks below. Support varies by model (for example, audio and decision models run on fewer frameworks), so check each model page.
- Transformers - For research and development
- llama.cpp - For efficient CPU inference, including iOS and Android apps
- vLLM - For high-throughput production serving
- zml/llmd - For high-throughput and fast startup
- MLX - For Apple Silicon optimization
- Ollama - For easy local deployment
Model Selection
Which model should I use for my use case?
Which model should I use for my use case?
- General chat/instruction following: LFM3-1.7B
- Vision tasks: LFM3-VL-2B
- Audio/speech: LFM3-Audio-2B
- Combined vision and audio: LFM3-Omni-2B
- Extraction from text: LFM3-1.7B with structured output
- Classification, routing, and scoring: decision models such as d1-2B
- Specialized extraction from images: LFM2.5-VL-1.6B-Extract or LFM2.5-VL-450M-Extract
- Smallest footprint: LFM2.5-230M or LFM2.5-350M for tight memory budgets
- Larger reasoning model: LFM2.5-24B-A2B (MoE architecture, 131K context); evaluate it on your task when a larger model fits your hardware
What is the difference between LFM versions?
What is the difference between LFM versions?
- LFM3 is the latest generation. Its tri-hybrid architecture combines gated short convolutions, sliding-window attention, and global attention. The family includes text, vision, audio, omni, and decision models, with updated chat templates and tool-call formats.
- LFM2.5 builds on the LFM2 architecture with improved training. Some models extend the context window or vocabulary. It remains useful for smaller footprints, larger MoE models, and specialized tasks.
- LFM2 is the earlier generation. Some specialized models remain useful; deprecated models and their replacements are listed on the Deprecations page.
What are Liquid Nanos?
What are Liquid Nanos?
Liquid Nanos are task-specific models fine-tuned for specialized use cases like:
- Structured extraction from images (LFM2.5-VL-1.6B-Extract, LFM2.5-VL-450M-Extract)
- Retrieval for RAG pipelines (LFM2.5-Embedding-350M, LFM2.5-ColBERT-350M)
- Translation (LFM2-350M-ENJP-MT)
- Meeting summarization (LFM2-2.6B-Transcript)
Deployment
Can I run LFM models on mobile devices?
Can I run LFM models on mobile devices?
Yes! llama.cpp runs natively on iOS and Android with quantized GGUF models. See iOS & Android for embedding it in an app, and Vision & Audio for multimodal models.
What quantization formats are available?
What quantization formats are available?
- GGUF: For llama.cpp, LM Studio, Ollama, Atomic Chat (Q4_0, Q4_K_M, Q5_K_M, Q6_K, Q8_0, BF16, F16; some repositories ship a subset)
- MLX: For Apple Silicon (4-bit, 5-bit, 6-bit, 8-bit, bf16)
- ONNX: For cross-platform deployment with ONNX Runtime
How do I choose the right quantization level?
How do I choose the right quantization level?
- Q4_0 / 4-bit: Smallest size, fastest inference, some quality loss
- Q8_0 / 8-bit: Good balance of size and quality
- F16 / bf16: Full precision, best quality, largest size
Fine-tuning
Can I fine-tune LFM models?
Can I fine-tune LFM models?
Yes. Most open-weight LFM models support fine-tuning with TRL and Unsloth, MoE models can be trained across GPUs with Halo, and audio models use the liquid-audio trainer. Check the Model Library for trainability information.
What fine-tuning methods are supported?
What fine-tuning methods are supported?
- LoRA/QLoRA: Memory-efficient fine-tuning
- Full fine-tuning: For maximum customization
- SFT (Supervised Fine-Tuning): For instruction tuning
Still Have Questions?
- Join our Discord community for real-time help
- Check the Cookbook for examples
- See Troubleshooting for common issues