Llamacpp imatrix Quantizations of miniclaus-qw1.5B-UNAMGS

Using llama.cpp release b4058 for quantization.

Original model: https://huggingface.co/fblgit/miniclaus-qw1.5B-UNAMGS

Trained with Magpie-Align/Magpie-Pro-MT-300K-v0.1

Using MGS & UNA (MLP) on this tiny but powerful model.

miniclaus-qw1.5B-UNAMGS Built with Axolotl

It achieves the following results on the evaluation set:

  • Loss: 0.7193

Training procedure

Click here for training details

Training hyperparameters

The following hyperparameters were used during training:

  • train_batch_size: 1
  • seed: 42
  • distributed_type: multi-GPU
  • num_devices: 8
  • total_train_batch_size: 128
  • total_eval_batch_size: 8
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • num_epochs: 1

Training results

Training Loss Epoch Step Validation Loss
1.1641 0.0007 1 0.8514
0.9246 0.0503 76 0.7921
0.8791 0.1006 152 0.7727
0.8507 0.1509 228 0.7611
0.8376 0.2012 304 0.7534
0.793 0.2515 380 0.7467
0.7834 0.3018 456 0.7421
0.7807 0.3521 532 0.7384
0.764 0.4023 608 0.7359
0.7738 0.4526 684 0.7320
0.7425 0.5029 760 0.7300
0.7519 0.5532 836 0.7279
0.7461 0.6035 912 0.7255
0.7489 0.6538 988 0.7245
0.7614 0.7041 1064 0.7222
0.7576 0.7544 1140 0.7222
0.7303 0.8047 1216 0.7209
0.7332 0.8550 1292 0.7199
0.7541 0.9053 1368 0.7202
0.7369 0.9556 1444 0.7193

All quants made using imatrix option with dataset from here

Run them in LM Studio

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 Split Description
miniclaus-qw1.5B-UNAMGS-f16.gguf f16 3.56GB false Full F16 weights.
miniclaus-qw1.5B-UNAMGS-Q8_0.gguf Q8_0 1.89GB false Extremely high quality, generally unneeded but max available quant.
miniclaus-qw1.5B-UNAMGS-Q6_K_L.gguf Q6_K_L 1.58GB false Uses Q8_0 for embed and output weights. Very high quality, near perfect, recommended.
miniclaus-qw1.5B-UNAMGS-Q6_K.gguf Q6_K 1.46GB false Very high quality, near perfect, recommended.
miniclaus-qw1.5B-UNAMGS-Q5_K_L.gguf Q5_K_L 1.43GB false Uses Q8_0 for embed and output weights. High quality, recommended.
miniclaus-qw1.5B-UNAMGS-Q5_K_M.gguf Q5_K_M 1.29GB false High quality, recommended.
miniclaus-qw1.5B-UNAMGS-Q4_K_L.gguf Q4_K_L 1.29GB false Uses Q8_0 for embed and output weights. Good quality, recommended.
miniclaus-qw1.5B-UNAMGS-Q5_K_S.gguf Q5_K_S 1.26GB false High quality, recommended.
miniclaus-qw1.5B-UNAMGS-Q3_K_XL.gguf Q3_K_XL 1.18GB false Uses Q8_0 for embed and output weights. Lower quality but usable, good for low RAM availability.
miniclaus-qw1.5B-UNAMGS-Q4_K_M.gguf Q4_K_M 1.12GB false Good quality, default size for most use cases, recommended.
miniclaus-qw1.5B-UNAMGS-Q4_K_S.gguf Q4_K_S 1.07GB false Slightly lower quality with more space savings, recommended.
miniclaus-qw1.5B-UNAMGS-Q4_0_8_8.gguf Q4_0_8_8 1.07GB false Optimized for ARM inference. Requires 'sve' support (see link below). Don't use on Mac or Windows.
miniclaus-qw1.5B-UNAMGS-Q4_0_4_8.gguf Q4_0_4_8 1.07GB false Optimized for ARM inference. Requires 'i8mm' support (see link below). Don't use on Mac or Windows.
miniclaus-qw1.5B-UNAMGS-Q4_0_4_4.gguf Q4_0_4_4 1.07GB false Optimized for ARM inference. Should work well on all ARM chips, pick this if you're unsure. Don't use on Mac or Windows.
miniclaus-qw1.5B-UNAMGS-Q4_0.gguf Q4_0 1.07GB false Legacy format, generally not worth using over similarly sized formats
miniclaus-qw1.5B-UNAMGS-IQ4_XS.gguf IQ4_XS 1.02GB false Decent quality, smaller than Q4_K_S with similar performance, recommended.
miniclaus-qw1.5B-UNAMGS-Q3_K_L.gguf Q3_K_L 0.98GB false Lower quality but usable, good for low RAM availability.
miniclaus-qw1.5B-UNAMGS-Q2_K_L.gguf Q2_K_L 0.98GB false Uses Q8_0 for embed and output weights. Very low quality but surprisingly usable.
miniclaus-qw1.5B-UNAMGS-Q3_K_M.gguf Q3_K_M 0.92GB false Low quality.
miniclaus-qw1.5B-UNAMGS-IQ3_M.gguf IQ3_M 0.88GB false Medium-low quality, new method with decent performance comparable to Q3_K_M.
miniclaus-qw1.5B-UNAMGS-Q3_K_S.gguf Q3_K_S 0.86GB false Low quality, not recommended.
miniclaus-qw1.5B-UNAMGS-IQ3_XS.gguf IQ3_XS 0.83GB false Lower quality, new method with decent performance, slightly better than Q3_K_S.
miniclaus-qw1.5B-UNAMGS-Q2_K.gguf Q2_K 0.75GB false Very low quality but surprisingly usable.
miniclaus-qw1.5B-UNAMGS-IQ2_M.gguf IQ2_M 0.70GB false Relatively low quality, uses SOTA techniques to be surprisingly usable.

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.

Some say that this improves the quality, others don't notice any difference. If you use these models PLEASE COMMENT with your findings. I would like feedback that these are actually used and useful so I don't keep uploading quants no one is using.

Thanks!

Downloading using huggingface-cli

First, make sure you have hugginface-cli installed:

pip install -U "huggingface_hub[cli]"

Then, you can target the specific file you want:

huggingface-cli download bartowski/miniclaus-qw1.5B-UNAMGS-GGUF --include "miniclaus-qw1.5B-UNAMGS-Q4_K_M.gguf" --local-dir ./

If the model is bigger than 50GB, it will have been split into multiple files. In order to download them all to a local folder, run:

huggingface-cli download bartowski/miniclaus-qw1.5B-UNAMGS-GGUF --include "miniclaus-qw1.5B-UNAMGS-Q8_0/*" --local-dir ./

You can either specify a new local-dir (miniclaus-qw1.5B-UNAMGS-Q8_0) or download them all in place (./)

Q4_0_X_X

These are NOT for Metal (Apple) offloading, only ARM chips.

If you're using an ARM chip, the Q4_0_X_X quants will have a substantial speedup. Check out Q4_0_4_4 speed comparisons on the original pull request

To check which one would work best for your ARM chip, you can check AArch64 SoC features (thanks EloyOn!).

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.

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

Downloads last month
1,209
GGUF
Model size
1.78B params
Architecture
qwen2

2-bit

3-bit

4-bit

5-bit

6-bit

8-bit

16-bit

Inference Examples
Unable to determine this model's library. Check the docs .

Model tree for bartowski/miniclaus-qw1.5B-UNAMGS-GGUF

Base model

Qwen/Qwen2.5-1.5B
Quantized
(2)
this model

Dataset used to train bartowski/miniclaus-qw1.5B-UNAMGS-GGUF