FAQs

Frequently asked questions about LFM models and deployment.

General Questions

What are LFM models?

LFM (Liquid Foundation Models) are a family of efficient language models built on a new hybrid architecture designed for fast training and inference. They range from 350M to 8B parameters and support text, vision, and audio modalities.

What context length do LFM models support?

Most LFM models support a 32K token context length for extended conversations and document processing. The LFM2.5-8B-A1B supports a 128K token context length.

Which inference frameworks are supported?

LFM models are compatible with:

• Transformers - For research and development
• llama.cpp - For efficient CPU inference
• vLLM - For high-throughput production serving
• MLX - For Apple Silicon optimization
• Ollama - For easy local deployment
• LEAP - For edge and mobile deployment

Model Selection

Which model should I use for my use case?

• General chat/instruction following: LFM2.5-1.2B-Instruct (recommended)
• Vision tasks: LFM2.5-VL-1.6B
• Audio/speech: LFM2.5-Audio-1.5B
• Extraction tasks: LFM2-1.2B-Extract or LFM2-350M-Extract
• Edge deployment: LFM2-350M or LFM2-700M for smallest footprint
• Highest performance: LFM2.5-8B-A1B (MoE architecture, 128K context)

What is the difference between LFM2 and LFM2.5?

LFM2.5 models are updated versions with improved training that deliver higher performance while maintaining the same architecture. We recommend using LFM2.5 variants when available.

What are Liquid Nanos?

Liquid Nanos are task-specific models fine-tuned for specialized use cases like:

• Information extraction (LFM2-Extract)
• Translation (LFM2-350M-ENJP-MT)
• RAG question answering (LFM2-1.2B-RAG)
• Meeting summarization (LFM2-2.6B-Transcript)

Deployment

Can I run LFM models on mobile devices?

Yes! Use the LEAP SDK to deploy models on iOS and Android devices. LEAP provides optimized inference for edge deployment with support for quantized models.

What quantization formats are available?

• GGUF: For llama.cpp, LM Studio, Ollama (Q4_0, Q4_K_M, Q5_K_M, Q6_K, Q8_0, F16)
• 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?

• 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

For most use cases, Q4_K_M or Q5_K_M provide good quality with significant size reduction.

Fine-tuning

Can I fine-tune LFM models?

Yes! Most LFM models support fine-tuning with TRL and Unsloth. Check the Model Library for trainability information.

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