LFM2.5-2.6B GGUF size and VRAM requirements

License: other ⬇ 31,489 ❤ 148
Parameters2.7B
Context128,000

LiquidAI/LFM2.5-2.6B-GGUF is a mid-size language model with 2.7 billion parameters, built on the lfm2 architecture. It is released under the other license and has been downloaded 31,489 times.

Under the hood it uses 30 transformer layers, a hidden size of 2,048, 32 attention heads. It uses a hybrid attention design: only 8 of its 30 layers keep a full KV cache, while the rest use linear attention with a constant-size state. Long contexts therefore cost far less memory than on a conventional model of this size.

To run LiquidAI/LFM2.5-2.6B-GGUF locally at a 4,096-token context, its quantized versions need between 2.35 GB (Q4_0, lowest quality) and 5.89 GB (BF16, highest quality) of memory, weights plus KV cache and a system margin included. Context length drives that memory directly: at 4,096 tokens the KV cache for LFM2.5-2.6B-GGUF is about 0.06 GB, rising to roughly 1.95 GB at its full 128,000-token context. Shorter prompts free up memory for a higher-quality quantization.

For most users the best balance is BF16, needing about 5.89 GB. That means LiquidAI/LFM2.5-2.6B-GGUF fits entirely in the VRAM of a 6 GB GPU or larger, running fully on the GPU.

Available GGUF quantizations for LiquidAI/LFM2.5-2.6B-GGUF include Q4_0, Q4_K_M, Q5_K_M, Q6_K, Q8_0, BF16, F16. The model supports a native context length of up to 128,000 tokens; a longer context grows the KV cache and the memory needed.

→ Guide: How much VRAM do you need?

GGUF file size and memory by quantization

Compare real GGUF weight sizes, estimated KV cache and total memory for Q4, Q5, Q8 and every quantization published in this repository.

Quant.Bits QualityWeights KVTotal Speed~Verdict
Q4_0 4.73 Good 1.48 GB 0.06 GB 2.35 GB 269.5 t/s Fits in VRAM
Q4_K_M 4.97 Good 1.56 GB 0.06 GB 2.42 GB 256.5 t/s Fits in VRAM
Q5_K_M 5.75 Very good 1.81 GB 0.06 GB 2.67 GB 221.4 t/s Fits in VRAM
Q6_K 6.59 Excellent 2.07 GB 0.06 GB 2.93 GB 193.3 t/s Fits in VRAM
Q8_0 8.53 Excellent 2.68 GB 0.06 GB 3.54 GB 149.4 t/s Fits in VRAM
BF16 16.03 Excellent 5.03 GB 0.06 GB 5.89 GB 79.5 t/s Fits in VRAM
F16 16.03 Excellent 5.03 GB 0.06 GB 5.89 GB 79.5 t/s Fits in VRAM

KV cache computed from the model's exact architecture. Speed is a rough estimate bounded by memory bandwidth.

Frequently asked questions

What kind of model is LiquidAI/LFM2.5-2.6B-GGUF?

LiquidAI/LFM2.5-2.6B-GGUF is a language model with 2.7 billion parameters, based on the lfm2 architecture. It is released under the other license and distributed as GGUF files for local inference.

How does LiquidAI/LFM2.5-2.6B-GGUF handle long context?

LiquidAI/LFM2.5-2.6B-GGUF uses a hybrid attention design: only 8 of its 30 layers keep a full KV cache, and the rest use linear attention with a constant-size state. Long contexts cost far less memory than on a conventional model of this size.

How much VRAM do you need to run LiquidAI/LFM2.5-2.6B-GGUF?

You need about 5.89 GB of VRAM to run LiquidAI/LFM2.5-2.6B-GGUF entirely on the GPU using the BF16 quantization (at a 4,096-token context). Smaller quantizations lower the requirement at the cost of quality.

Can I run LiquidAI/LFM2.5-2.6B-GGUF on an 8 GB GPU?

Yes. With 8 GB of VRAM you can run LiquidAI/LFM2.5-2.6B-GGUF fully on the GPU using BF16 (about 5.89 GB).

Can I run LiquidAI/LFM2.5-2.6B-GGUF on a 16 GB GPU?

Yes. With 16 GB of VRAM you can run LiquidAI/LFM2.5-2.6B-GGUF fully on the GPU using BF16 (about 5.89 GB).

Can I run LiquidAI/LFM2.5-2.6B-GGUF on a 24 GB GPU?

Yes. With 24 GB of VRAM you can run LiquidAI/LFM2.5-2.6B-GGUF fully on the GPU using BF16 (about 5.89 GB).

What context length does LiquidAI/LFM2.5-2.6B-GGUF support?

LiquidAI/LFM2.5-2.6B-GGUF supports a native context length of up to 128,000 tokens. A longer context grows the KV cache, so it increases the memory needed to run the model.

What is the best quantization for LiquidAI/LFM2.5-2.6B-GGUF?

For LiquidAI/LFM2.5-2.6B-GGUF, a strong default is Q4_K_M, which needs about 2.42 GB and keeps most of the quality while roughly halving the memory versus 8-bit. With VRAM to spare, Q5_K_M or Q6_K add a little more quality; if you are tight on memory, a smaller quantization still runs. Pick the highest quantization that fits your VRAM.