---
library_name: transformers
tags: []
quantized_by: bartowski
pipeline_tag: text-generation
---

## Llamacpp Quantizations of mamba-2.8b-hf

Using <a href="https://github.com/ggerganov/llama.cpp/">llama.cpp</a> release <a href="https://github.com/ggerganov/llama.cpp/releases/tag/b2536">b2536</a> for quantization.

Original model: https://huggingface.co/state-spaces/mamba-2.8b-hf

Download a file (not the whole branch) from below:

| Filename | Quant type | File Size | Description |
| -------- | ---------- | --------- | ----------- |
| [mamba-2.8b-hf-Q8_0.gguf](https://huggingface.co/bartowski/mamba-2.8b-hf-GGUF/blob/main/mamba-2.8b-hf-Q8_0.gguf) | Q8_0 | 3.30GB | Extremely high quality, generally unneeded but max available quant. |
| [mamba-2.8b-hf-Q6_K.gguf](https://huggingface.co/bartowski/mamba-2.8b-hf-GGUF/blob/main/mamba-2.8b-hf-Q6_K.gguf) | Q6_K | 2.66GB | Very high quality, near perfect, *recommended*. |
| [mamba-2.8b-hf-Q5_K_M.gguf](https://huggingface.co/bartowski/mamba-2.8b-hf-GGUF/blob/main/mamba-2.8b-hf-Q5_K_M.gguf) | Q5_K_M | 2.32GB | High quality, very usable. |
| [mamba-2.8b-hf-Q5_K_S.gguf](https://huggingface.co/bartowski/mamba-2.8b-hf-GGUF/blob/main/mamba-2.8b-hf-Q5_K_S.gguf) | Q5_K_S | 2.32GB | High quality, very usable. |
| [mamba-2.8b-hf-Q5_0.gguf](https://huggingface.co/bartowski/mamba-2.8b-hf-GGUF/blob/main/mamba-2.8b-hf-Q5_0.gguf) | Q5_0 | 2.32GB | High quality, older format, generally not recommended. |
| [mamba-2.8b-hf-Q4_K_M.gguf](https://huggingface.co/bartowski/mamba-2.8b-hf-GGUF/blob/main/mamba-2.8b-hf-Q4_K_M.gguf) | Q4_K_M | 2.01GB | Good quality, uses about 4.83 bits per weight. |
| [mamba-2.8b-hf-Q4_K_S.gguf](https://huggingface.co/bartowski/mamba-2.8b-hf-GGUF/blob/main/mamba-2.8b-hf-Q4_K_S.gguf) | Q4_K_S | 2.01GB | Slightly lower quality with small space savings. |
| [mamba-2.8b-hf-IQ4_NL.gguf](https://huggingface.co/bartowski/mamba-2.8b-hf-GGUF/blob/main/mamba-2.8b-hf-IQ4_NL.gguf) | IQ4_NL | 2.01GB | Decent quality, similar to Q4_K_S, new method of quanting, |
| [mamba-2.8b-hf-IQ4_XS.gguf](https://huggingface.co/bartowski/mamba-2.8b-hf-GGUF/blob/main/mamba-2.8b-hf-IQ4_XS.gguf) | IQ4_XS | 1.93GB | Decent quality, new method with similar performance to Q4. |
| [mamba-2.8b-hf-Q4_0.gguf](https://huggingface.co/bartowski/mamba-2.8b-hf-GGUF/blob/main/mamba-2.8b-hf-Q4_0.gguf) | Q4_0 | 2.01GB | Decent quality, older format, generally not recommended. |
| [mamba-2.8b-hf-Q3_K_L.gguf](https://huggingface.co/bartowski/mamba-2.8b-hf-GGUF/blob/main/mamba-2.8b-hf-Q3_K_L.gguf) | Q3_K_L | 1.68GB | Lower quality but usable, good for low RAM availability. |
| [mamba-2.8b-hf-Q3_K_M.gguf](https://huggingface.co/bartowski/mamba-2.8b-hf-GGUF/blob/main/mamba-2.8b-hf-Q3_K_M.gguf) | Q3_K_M | 1.68GB | Even lower quality. |
| [mamba-2.8b-hf-IQ3_M.gguf](https://huggingface.co/bartowski/mamba-2.8b-hf-GGUF/blob/main/mamba-2.8b-hf-IQ3_M.gguf) | IQ3_M | 1.68GB | Medium-low quality, new method with decent performance. |
| [mamba-2.8b-hf-IQ3_S.gguf](https://huggingface.co/bartowski/mamba-2.8b-hf-GGUF/blob/main/mamba-2.8b-hf-IQ3_S.gguf) | IQ3_S | 1.68GB | Lower quality, new method with decent performance, recommended over Q3 quants. |
| [mamba-2.8b-hf-Q3_K_S.gguf](https://huggingface.co/bartowski/mamba-2.8b-hf-GGUF/blob/main/mamba-2.8b-hf-Q3_K_S.gguf) | Q3_K_S | 1.68GB | Low quality, not recommended. |
| [mamba-2.8b-hf-Q2_K.gguf](https://huggingface.co/bartowski/mamba-2.8b-hf-GGUF/blob/main/mamba-2.8b-hf-Q2_K.gguf) | Q2_K | 1.42GB | Extremely low quality, *not* recommended.

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