LFM2.5 8B-A1B

Running LFM2.5 8B-A1B locally

At Q4_K_M and 4K of context, LFM2.5 8B-A1B needs about 5.4 GB — 4.9 GB of weights, 0.05 GB of KV cache and a 0.5 GB activation buffer. The smallest cards in this index that clear that comfortably are 8 GB ones — 9 of them, from the RTX 4060 up — and 76 of the 78 cards here do. These are calculated figures, not measurements: the estimate stops counting a card as comfortable at 88% of its VRAM, which is roughly the room a desktop session needs.

What longer context costs

Going from 4K to 32K adds about 0.33 GB, taking the total to 5.8 GB. That is gentler than the parameter count suggests: only 6 of 24 layers keep a KV cache that grows with context, so the cache scales at a fraction of the usual rate. The model's native window is 125K; holding all of it at Q4_K_M would need about 7 GB.

Which build to download

This index only tracks LFM2.5 8B-A1B in GGUF, across 3 levels (Q4_K_M, Q5_K_M, Q8_0). No per-level perplexity sweep has been published for this model, so the quality column is empty rather than estimated — within one model, more bits per weight is the only ordering the data supports. No throughput has been measured for this model on this site's hardware. GGUF runs on llama.cpp and Ollama across NVIDIA, AMD and Apple silicon; AWQ and GPTQ target vLLM on CUDA and ROCm; EXL2 is ExLlamaV2 — CUDA only, and now archived, so local use rather than serving.

Common questions

How much VRAM does LFM2.5 8B-A1B need?

About 5.4 GB at Q4_K_M with 4K of context and batch 1, rising to roughly 5.8 GB at 32K. That figure is weights plus KV cache plus a 10% activation buffer, calculated from the model's architecture rather than measured on a card.

Will LFM2.5 8B-A1B run on a 8GB GPU?

Yes — at Q4_K_M and 4K context it needs about 5.4 GB, which leaves 2.6 GB spare on a RTX 4060, or on any of the other 8 GB cards here (9 in all). That is the smallest size in this index that clears it comfortably; 76 of 78 do.

Which quantization of LFM2.5 8B-A1B should I use?

No published quality comparison exists for this model, so pick by footprint: Q4_K_M is the level most people run, and a higher bits-per-weight level is more faithful. All 3 levels in the index are Q4_K_M, Q5_K_M, Q8_0. Note this is a mixture-of-experts model — all parameters must be resident even though only a fraction are active per token, so the memory cost follows the total, not the active count.