Model family · 7 sizes
LFM: which size runs locally?
LFM comes in 7 sizes, from 0.354B to 24B. The LFM2 350M is small enough for a phone, while the LFM2 24B-A2B wants a workstation. Here is each size with its Q4_K_M weight, the memory it needs, and the hardware that runs it.
- Sizes
- 7
- Smallest
- 0.354B
- Largest
- 24B
- Runs from
- 8GB
The LFM lineup
"Needs" is the sourced minimum memory for Q4_K_M with a small context. Larger context needs more.
Which LFM fits your memory
Largest that fits: LFM2.5 1.2B Thinking (1.17B), best case on Apple M1 (8GB).
Largest that fits: LFM2.5 8B-A1B (8.3B), best case on Nvidia GeForce RTX 4080 (16GB).
Largest that fits: LFM2 24B-A2B (24B), best case on Nvidia GeForce RTX 4090 (24GB).
Largest that fits: LFM2 24B-A2B (24B), best case on Nvidia GeForce RTX 5090 (32GB).
Largest that fits: LFM2 24B-A2B (24B), best case on Apple M5 Pro (48GB).
Largest that fits: LFM2 24B-A2B (24B), best case on Apple M4 Max (64GB).
Largest that fits: LFM2 24B-A2B (24B), best case on Apple M5 Max (128GB).
Largest that fits: LFM2 24B-A2B (24B), best case on Apple M3 Ultra (256GB).
Best case means the most capable device at that size (usually a discrete GPU). A Mac at the same size sits roughly one rung lower; see the per-size breakdown on each memory budget page.
FAQ
Which LFM size should I run locally?
Pick the largest size your memory allows. On 8GB (best case) up to LFM2.5 1.2B Thinking; On 16GB (best case) up to LFM2.5 8B-A1B; On 24GB (best case) up to LFM2 24B-A2B; On 32GB (best case) up to LFM2 24B-A2B; On 48GB (best case) up to LFM2 24B-A2B; On 64GB (best case) up to LFM2 24B-A2B; On 128GB (best case) up to LFM2 24B-A2B; On 256GB (best case) up to LFM2 24B-A2B. Smaller sizes run faster and leave headroom for context.
What is the smallest LFM model?
LFM2 350M at 0.354B parameters, about 0.45 GB on disk at Q4_K_M. It is the one to use on phones and 8 GB machines.
What is the largest LFM model and what does it need?
LFM2 24B-A2B at 24B (mixture of experts), about 13.43 GB at Q4_K_M. It fits a high-memory desktop GPU or Mac.
Understand the numbers
Short guides to the ideas behind LFM's memory and quant figures.
Sources
- arxiv.org
- huggingface.co/LiquidAI/LFM2-1.2B
- huggingface.co/LiquidAI/LFM2-1.2B-GGUF
- huggingface.co/LiquidAI/LFM2-350M
- huggingface.co/LiquidAI/LFM2-350M-GGUF
- huggingface.co/LiquidAI/LFM2-700M
- huggingface.co/LiquidAI/LFM2-700M-GGUF
- huggingface.co/LiquidAI/LFM2.5-1.2B-Instruct
- huggingface.co/LiquidAI/LFM2.5-1.2B-Instruct-GGUF
- huggingface.co/LiquidAI/LFM2.5-1.2B-Thinking
- huggingface.co/LiquidAI/LFM2.5-1.2B-Thinking-GGUF
- huggingface.co/LiquidAI/LFM2.5-8B-A1B
- huggingface.co/LiquidAI/LFM2.5-8B-A1B-GGUF
- liquid.ai
- ollama.com/library/lfm2.5
- ollama.com/library/lfm2.5-thinking
Memory figures are estimates at Q4_K_M. See methodology.