Edit model card

Llamacpp imatrix Quantizations of Qwen1.5-110B-Chat

Using llama.cpp release b2717 for quantization.

Original model: https://huggingface.co/Qwen/Qwen1.5-110B-Chat

All quants made using imatrix option with dataset provided by Kalomaze here

Prompt format

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

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

Filename Quant type File Size Description
Qwen1.5-110B-Chat-Q8_0.gguf Q8_0 118.17GB Extremely high quality, generally unneeded but max available quant.
Qwen1.5-110B-Chat-Q6_K.gguf Q6_K 91.23GB Very high quality, near perfect, recommended.
Qwen1.5-110B-Chat-Q5_K_M.gguf Q5_K_M 78.81GB High quality, recommended.
Qwen1.5-110B-Chat-Q5_K_S.gguf Q5_K_S 76.63GB High quality, recommended.
Qwen1.5-110B-Chat-Q4_K_M.gguf Q4_K_M 67.17GB Good quality, uses about 4.83 bits per weight, recommended.
Qwen1.5-110B-Chat-Q4_K_S.gguf Q4_K_S 63.47GB Slightly lower quality with more space savings, recommended.
Qwen1.5-110B-Chat-IQ4_NL.gguf IQ4_NL 62.97GB Decent quality, slightly smaller than Q4_K_S with similar performance recommended.
Qwen1.5-110B-Chat-IQ4_XS.gguf IQ4_XS 59.55GB Decent quality, smaller than Q4_K_S with similar performance, recommended.
Qwen1.5-110B-Chat-Q3_K_L.gguf Q3_K_L 58.14GB Lower quality but usable, good for low RAM availability.
Qwen1.5-110B-Chat-Q3_K_M.gguf Q3_K_M 53.70GB Even lower quality.
Qwen1.5-110B-Chat-IQ3_M.gguf IQ3_M 49.70GB Medium-low quality, new method with decent performance comparable to Q3_K_M.
Qwen1.5-110B-Chat-IQ3_S.gguf IQ3_S 48.45GB Lower quality, new method with decent performance, recommended over Q3_K_S quant, same size with better performance.
Qwen1.5-110B-Chat-Q3_K_S.gguf Q3_K_S 48.45GB Low quality, not recommended.
Qwen1.5-110B-Chat-IQ3_XS.gguf IQ3_XS 45.91GB Lower quality, new method with decent performance, slightly better than Q3_K_S.
Qwen1.5-110B-Chat-IQ3_XXS.gguf IQ3_XXS 43.10GB Lower quality, new method with decent performance, comparable to Q3 quants.
Qwen1.5-110B-Chat-Q2_K.gguf Q2_K 41.17GB Very low quality but surprisingly usable.
Qwen1.5-110B-Chat-IQ2_M.gguf IQ2_M 37.41GB Very low quality, uses SOTA techniques to also be surprisingly usable.
Qwen1.5-110B-Chat-IQ2_S.gguf IQ2_S 34.33GB Very low quality, uses SOTA techniques to be usable.
Qwen1.5-110B-Chat-IQ2_XS.gguf IQ2_XS 33.04GB Very low quality, uses SOTA techniques to be usable.
Qwen1.5-110B-Chat-IQ2_XXS.gguf IQ2_XXS 29.79GB Lower quality, uses SOTA techniques to be usable.
Qwen1.5-110B-Chat-IQ1_M.gguf IQ1_M 25.94GB Extremely low quality, not recommended.
Qwen1.5-110B-Chat-IQ1_S.gguf IQ1_S 23.63GB Extremely low quality, not recommended.

Which file should I choose?

A great write up with charts showing various 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.

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 and Apple Metal, but will be slower than their K-quant equivalent, so speed vs performance is a tradeoff you'll have to decide.

The I-quants are not compatible with Vulcan, which is also AMD, so if you have an AMD card double check if you're using the rocBLAS build or the Vulcan build. At the time of writing this, LM Studio has a preview with ROCm support, and other inference engines have specific builds for ROCm.

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

Downloads last month
1,971
GGUF
Model size
111B params
Architecture
qwen2
+14
Inference Examples
Unable to determine this model's library. Check the docs .