LFM2-350M-Extract

10 of 16 layers use linear attention, which keeps a fixed-size recurrent state instead of a per-token cache. Those layers do not grow with context at all — treating them as ordinary attention, as a flat formula does, overstates this model's cache by roughly 2.7× at long context.

A parameters × bits ÷ 8 estimate puts Q4_K_M at roughly 0.19 GiB. The real file is 0.21 GiB, because a quantization is a mixture and some tensors are always kept at higher precision.

How much VRAM does LFM2-350M-Extract need?

Q4_K_M is exactly 229,310,080 bytes (0.21 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.

How large is LFM2-350M-Extract's KV cache?

0.38 GiB at 32K context with an f16 cache, computed per layer. Quantizing the cache to q8_0 roughly halves it, which is often the difference between a context length fitting and not.

Which quantization of LFM2-350M-Extract should I use?

Q4_K_M is the usual default. Pick the largest quantization that fits your card at the context you actually need — the table above gives exact sizes for every one published.