bartowski/Naphula_Cyberdonia-24B-v8-GGUF

Llamacpp imatrix Quantizations of Cyberdonia-24B-v8 by Naphula

Using llama.cpp release b11845 for quantization.

Original model: https://huggingface.co/Naphula/Cyberdonia-24B-v8

Model details:

• Parameter count: 24B (source checkpoint)
• Input support: text
• Speculative decoding: no
• imatrix: yes - details

How to run

Prompt format

<s>[SYSTEM_PROMPT]{system_prompt}[/SYSTEM_PROMPT][INST]{prompt}[/INST]

Don't know which to choose? Grab Q4_K_M (14.57GB) - usually a good mix of size and performance. Download instructions available here

Available files:

Filename	Quant type	File Size	Split	Description
Naphula_Cyberdonia-24B-v8-bf16.gguf	bf16	47.15GB	false	Full BF16 weights.
Naphula_Cyberdonia-24B-v8-Q8_0.gguf	Q8_0	25.05GB	false	Extremely high quality, generally unneeded but max available quant.
Naphula_Cyberdonia-24B-v8-Q6_K_L.gguf	Q6_K_L	20.05GB	false	Uses Q8_0 for embed and output weights. Very high quality, near perfect, recommended.
Naphula_Cyberdonia-24B-v8-Q6_K.gguf	Q6_K	19.73GB	false	Very high quality, near perfect, recommended.
Naphula_Cyberdonia-24B-v8-Q5_K_M.gguf	Q5_K_M	16.96GB	false	High quality, recommended.
Naphula_Cyberdonia-24B-v8-Q5_K_S.gguf	Q5_K_S	16.36GB	false	High quality, recommended.
Naphula_Cyberdonia-24B-v8-Q4_K_L.gguf	Q4_K_L	15.07GB	false	Uses Q8_0 for embed and output weights. Good quality, recommended.
Naphula_Cyberdonia-24B-v8-Q4_1.gguf	Q4_1	14.96GB	false	Legacy format, similar performance to Q4_K_S but with improved tokens/watt on Apple silicon.
Naphula_Cyberdonia-24B-v8-Q4_K_M.gguf	Q4_K_M	14.57GB	false	Good quality, default size for most use cases, recommended.
Naphula_Cyberdonia-24B-v8-Q4_K_S.gguf	Q4_K_S	13.67GB	false	Slightly lower quality with more space savings, recommended.
Naphula_Cyberdonia-24B-v8-IQ4_NL.gguf	IQ4_NL	13.60GB	false	Similar to IQ4_XS, but slightly larger.
Naphula_Cyberdonia-24B-v8-Q4_0.gguf	Q4_0	13.60GB	false	Legacy format, kept for compatibility with older tools.
Naphula_Cyberdonia-24B-v8-IQ4_XS.gguf	IQ4_XS	12.91GB	false	Decent quality, smaller than Q4_K_S with similar performance, recommended.
Naphula_Cyberdonia-24B-v8-Q3_K_L.gguf	Q3_K_L	12.46GB	false	Lower quality but usable, good for low RAM availability.
Naphula_Cyberdonia-24B-v8-Q3_K_M.gguf	Q3_K_M	11.65GB	false	Low quality.
Naphula_Cyberdonia-24B-v8-IQ3_M.gguf	IQ3_M	10.84GB	false	Medium-low quality, new method with decent performance comparable to Q3_K_M.
Naphula_Cyberdonia-24B-v8-Q3_K_S.gguf	Q3_K_S	10.56GB	false	Low quality, not recommended.
Naphula_Cyberdonia-24B-v8-IQ3_XS.gguf	IQ3_XS	10.12GB	false	Lower quality, new method with decent performance, slightly better than Q3_K_S.
Naphula_Cyberdonia-24B-v8-IQ3_XXS.gguf	IQ3_XXS	9.53GB	false	Lower quality, new method with decent performance, comparable to Q3 quants.
Naphula_Cyberdonia-24B-v8-Q2_K.gguf	Q2_K	9.10GB	false	Very low quality but surprisingly usable.
Naphula_Cyberdonia-24B-v8-IQ2_M.gguf	IQ2_M	8.80GB	false	Relatively low quality, uses SOTA techniques to be surprisingly usable.
Naphula_Cyberdonia-24B-v8-IQ2_S.gguf	IQ2_S	8.25GB	false	Low quality, uses SOTA techniques to be usable.
Naphula_Cyberdonia-24B-v8-IQ2_XS.gguf	IQ2_XS	7.91GB	false	Low quality, uses SOTA techniques to be usable.

Download a specific file:

hf download bartowski/Naphula_Cyberdonia-24B-v8-GGUF --include "Naphula_Cyberdonia-24B-v8-Q4_K_M.gguf" --local-dir ./

Downloading using the Hugging Face CLI

Click to view download instructions

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

pip install -U "huggingface_hub[cli]"

Download a specific file:

hf download bartowski/Naphula_Cyberdonia-24B-v8-GGUF --include "Naphula_Cyberdonia-24B-v8-Q4_K_M.gguf" --local-dir ./

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/Naphula_Cyberdonia-24B-v8-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 b11845 - 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 a calibration corpus rendered through this model's own chat template. The corpus pairs plain prose with tool-calling and reasoning conversations (corpus source data), encoded exactly as this model sees them at inference and processed with --parse-special, so chat-format special tokens contribute to the importance matrix. The corpus rendered for this model is included in this repo: Naphula_Cyberdonia-24B-v8-calibration-v6.txt. The imatrix is available here: Naphula_Cyberdonia-24B-v8-imatrix.gguf.

Calibration render details

{
  "recipe": "calibration-v6",
  "encoder": "chat_template",
  "library_versions": {
    "transformers": "5.20.2",
    "tokenizers": "0.23.3",
    "tiktoken": "0.14.0",
    "blobfile": "3.3.0",
    "huggingface_hub": "1.31.0"
  },
  "chunk_size": 512,
  "prose_chunks": 222,
  "tool_chunks": 269,
  "total_chunks": 491,
  "n_conversations": 173,
  "conversation_token_lengths": [
    239,
    782,
    526,
    1153,
    757,
    594,
    661,
    2097,
    1068,
    357,
    570,
    915,
    770,
    1131,
    404,
    629,
    2095,
    750,
    660,
    558,
    619,
    286,
    233,
    558,
    637,
    559,
    1002,
    552,
    479,
    748,
    1508,
    746,
    592,
    428,
    581,
    1206,
    989,
    404,
    470,
    649,
    631,
    1280,
    1032,
    624,
    161,
    712,
    578,
    607,
    949,
    1531,
    878,
    1703,
    640,
    887,
    492,
    2300,
    2097,
    380,
    2393,
    435,
    637,
    415,
    510,
    296,
    1965,
    466,
    563,
    581,
    584,
    654,
    533,
    650,
    575,
    510,
    670,
    649,
    607,
    1080,
    340,
    756,
    1564,
    792,
    184,
    302,
    516,
    669,
    2836,
    2045,
    276,
    1214,
    420,
    579,
    456,
    685,
    243,
    664,
    366,
    759,
    582,
    838,
    983,
    702,
    620,
    853,
    774,
    483,
    573,
    311,
    422,
    1094,
    2158,
    2102,
    502,
    235,
    1251,
    804,
    816,
    2077,
    1342,
    541,
    234,
    511,
    1281,
    1055,
    1429,
    1112,
    490,
    604,
    466,
    159,
    596,
    2148,
    329,
    866,
    371,
    384,
    885,
    1014,
    169,
    1051,
    357,
    381,
    317,
    278,
    292,
    198,
    1978,
    656,
    410,
    613,
    496,
    1620,
    1370,
    2037,
    2078,
    422,
    584,
    900,
    416,
    290,
    477,
    593,
    528,
    1225,
    280,
    775,
    477,
    322,
    1060,
    583,
    440,
    703,
    1013
  ]
}

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?

Click here for details

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.

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