bartowski/LiquidAI_LFM2.5-2.6B-GGUF overview

---

quantized_by: bartowski

pipeline_tag: text-generation

license_link: LICENSE

language:

• ar
• zh
• en
• fr
• de
• hi
• id
• it
• ja
• ko
• pl
• pt
• ru
• es
• th
• vi

base_model_relation: quantized

base_model: LiquidAI/LFM2.5-2.6B

license_name: lfm1.0

tags:

• liquid
• lfm2.5
• edge

license: other

---

Llamacpp imatrix Quantizations of LFM2.5-2.6B by LiquidAI

Using <a href="https://github.com/ggml-org/llama.cpp/">llama.cpp</a> release <a href="https://github.com/ggml-org/llama.cpp/releases/tag/b10262">b10262</a> for quantization.

Original model: https://huggingface.co/LiquidAI/LFM2.5-2.6B

Model details:

• Parameter count: 3B
• Input support: text
• MTP: no
• imatrix: yes - details

How to run

Prompt format

<|startoftext|><|im_start|>system
{system_prompt}<|im_end|>
<|im_start|>user
{prompt}<|im_end|>
<|im_start|>assistant
<think>

Don't know which to choose? Grab Q4_K_M (1.68GB) - usually a good mix of size and performance. Download instructions available here

Available files:

| Filename | Quant type | File Size | Split | Description |

| -------- | ---------- | --------- | ----- | ----------- |

| LiquidAI_LFM2.5-2.6B-bf16.gguf | bf16 | 5.40GB | false | Full BF16 weights. |

| LiquidAI_LFM2.5-2.6B-Q8_0.gguf | Q8_0 | 2.87GB | false | Extremely high quality, generally unneeded but max available quant. |

| LiquidAI_LFM2.5-2.6B-Q6_K_L.gguf | Q6_K_L | 2.30GB | false | Uses Q8_0 for embed and output weights. Very high quality, near perfect, recommended. |

| LiquidAI_LFM2.5-2.6B-Q6_K.gguf | Q6_K | 2.24GB | false | Very high quality, near perfect, recommended. |

| LiquidAI_LFM2.5-2.6B-Q5_K_L.gguf | Q5_K_L | 2.01GB | false | Uses Q8_0 for embed and output weights. High quality, recommended. |

| LiquidAI_LFM2.5-2.6B-Q5_K_M.gguf | Q5_K_M | 1.95GB | false | High quality, recommended. |

| LiquidAI_LFM2.5-2.6B-Q5_K_S.gguf | Q5_K_S | 1.90GB | false | High quality, recommended. |

| LiquidAI_LFM2.5-2.6B-Q4_1.gguf | Q4_1 | 1.75GB | false | Legacy format, similar performance to Q4_K_S but with improved tokens/watt on Apple silicon. |

| LiquidAI_LFM2.5-2.6B-Q4_K_L.gguf | Q4_K_L | 1.75GB | false | Uses Q8_0 for embed and output weights. Good quality, recommended. |

| LiquidAI_LFM2.5-2.6B-Q4_K_M.gguf | Q4_K_M | 1.68GB | false | Good quality, default size for most use cases, recommended. |

| LiquidAI_LFM2.5-2.6B-Q4_K_S.gguf | Q4_K_S | 1.61GB | false | Slightly lower quality with more space savings, recommended. |

| LiquidAI_LFM2.5-2.6B-Q4_0.gguf | Q4_0 | 1.60GB | false | Legacy format, kept for compatibility with older tools. |

| LiquidAI_LFM2.5-2.6B-IQ4_NL.gguf | IQ4_NL | 1.60GB | false | Similar to IQ4_XS, but slightly larger. |

| LiquidAI_LFM2.5-2.6B-IQ4_XS.gguf | IQ4_XS | 1.52GB | false | Decent quality, smaller than Q4_K_S with similar performance, recommended. |

| LiquidAI_LFM2.5-2.6B-Q3_K_XL.gguf | Q3_K_XL | 1.52GB | false | Uses Q8_0 for embed and output weights. Lower quality but usable, good for low RAM availability. |

| LiquidAI_LFM2.5-2.6B-Q3_K_L.gguf | Q3_K_L | 1.45GB | false | Lower quality but usable, good for low RAM availability. |

| LiquidAI_LFM2.5-2.6B-Q3_K_M.gguf | Q3_K_M | 1.37GB | false | Low quality. |

| LiquidAI_LFM2.5-2.6B-IQ3_M.gguf | IQ3_M | 1.29GB | false | Medium-low quality, new method with decent performance comparable to Q3_K_M. |

| LiquidAI_LFM2.5-2.6B-Q3_K_S.gguf | Q3_K_S | 1.28GB | false | Low quality, not recommended. |

| LiquidAI_LFM2.5-2.6B-IQ3_XS.gguf | IQ3_XS | 1.23GB | false | Lower quality, new method with decent performance, slightly better than Q3_K_S. |

| LiquidAI_LFM2.5-2.6B-Q2_K_L.gguf | Q2_K_L | 1.17GB | false | Uses Q8_0 for embed and output weights. Very low quality but surprisingly usable. |

| LiquidAI_LFM2.5-2.6B-IQ3_XXS.gguf | IQ3_XXS | 1.13GB | false | Lower quality, new method with decent performance, comparable to Q3 quants. |

| LiquidAI_LFM2.5-2.6B-Q2_K.gguf | Q2_K | 1.10GB | false | Very low quality but surprisingly usable. |

| LiquidAI_LFM2.5-2.6B-IQ2_M.gguf | IQ2_M | 1.02GB | false | Relatively low quality, uses SOTA techniques to be surprisingly usable. |

Download a specific file:

hf download bartowski/LiquidAI_LFM2.5-2.6B-GGUF --include "LiquidAI_LFM2.5-2.6B-Q4_K_M.gguf" --local-dir ./

Downloading using the Hugging Face CLI

<details>

<summary>Click to view download instructions</summary>

First, make sure you have the Hugging Face CLI installed:

pip install -U "huggingface_hub[cli]"

Download a specific file:

hf download bartowski/LiquidAI_LFM2.5-2.6B-GGUF --include "LiquidAI_LFM2.5-2.6B-Q4_K_M.gguf" --local-dir ./

</details>

How to run

These quants run with llama.cpp - installable in one line via llama.app:

curl -LsSf https://llama.app/install.sh | sh
llama-server -hf bartowski/LiquidAI_LFM2.5-2.6B-GGUF:Q4_K_M

llama-server includes a built-in chat web UI, served at http://localhost:8080 by default.

These quants were made with llama.cpp release b10262 - if this model's architecture is newly supported, you'll need that release or newer to run them.

They also work in: LM Studio · koboldcpp · ramalama · Jan AI · Text Generation Web UI · LoLLMs · Atomic Chat

imatrix

All quants made using imatrix option with dataset from here. The imatrix is available here: LiquidAI_LFM2.5-2.6B-imatrix.gguf.

Embed/output weights

Some of these quants (Q3_K_XL, Q4_K_L etc) are the standard quantization method with the embeddings and output weights quantized to Q8_0 instead of what they would normally default to.

ARM/AVX information

llama.cpp automatically "repacks" weights into an interleaved layout at load time for faster inference on ARM and AVX machines - details in this PR. This once required downloading special Q4_0_4_4/4_8/8_8 files; those are long gone. Online repacking now covers Q4_0, IQ4_NL, and most K-quants, so no special quant choice is needed for CPU inference.

Which file should I choose?

<details>

<summary>Click here for details</summary>

An older (early 2024) but still useful write-up with charts comparing quant performances is provided by Artefact2 here

The first thing to figure out is how big a model you can run. To do this, you'll need to figure out how much RAM and/or VRAM you have.

If you want your model running as FAST as possible, you'll want to fit the whole thing on your GPU's VRAM. Aim for a quant with a file size 1-2GB smaller than your GPU's total VRAM.

If you want the absolute maximum quality, add both your system RAM and your GPU's VRAM together, then similarly grab a quant with a file size 1-2GB Smaller than that total.

Hugging Face can also do this math for you: add your hardware in your Local Apps settings and the model page will show which files fit.

Next, you'll need to decide if you want to use an 'I-quant' or a 'K-quant'.

If you don't want to think too much, grab one of the K-quants. These are in format 'QX_K_X', like Q5_K_M.

If you want to get more into the weeds, you can check out this extremely useful feature chart:

llama.cpp feature matrix

But basically, if you're aiming for below Q4, and you're running cuBLAS (Nvidia) or rocBLAS (AMD), you should look towards the I-quants. These are in format IQX_X, like IQ3_M. These are newer and offer better performance for their size.

These I-quants can also be used on CPU, but will be slower than their K-quant equivalent, so speed vs performance is a tradeoff you'll have to decide.

</details>

Credits

Thank you kalomaze and Dampf for assistance in creating the imatrix calibration dataset.

Thank you ZeroWw for the inspiration to experiment with embed/output.

Want to support my work? Visit my ko-fi page here: https://ko-fi.com/bartowski