VOOZH about

URL: https://huggingface.co/bartowski/FineMath-Llama-3B-GGUF

⇱ bartowski/FineMath-Llama-3B-GGUF · Hugging Face

Llamacpp imatrix Quantizations of FineMath-Llama-3B

Using llama.cpp release b4418 for quantization.

Original model: https://huggingface.co/HuggingFaceTB/FineMath-Llama-3B

All quants made using imatrix option with dataset from here

Run them in LM Studio

Prompt format

No prompt format found, check original model page

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

Filename Quant type File Size Split Description
FineMath-Llama-3B-f32.gguf f32 12.86GB false Full F32 weights.
FineMath-Llama-3B-f16.gguf f16 6.43GB false Full F16 weights.
FineMath-Llama-3B-Q8_0.gguf Q8_0 3.42GB false Extremely high quality, generally unneeded but max available quant.
FineMath-Llama-3B-Q6_K_L.gguf Q6_K_L 2.74GB false Uses Q8_0 for embed and output weights. Very high quality, near perfect, recommended.
FineMath-Llama-3B-Q6_K.gguf Q6_K 2.64GB false Very high quality, near perfect, recommended.
FineMath-Llama-3B-Q5_K_L.gguf Q5_K_L 2.42GB false Uses Q8_0 for embed and output weights. High quality, recommended.
FineMath-Llama-3B-Q5_K_M.gguf Q5_K_M 2.32GB false High quality, recommended.
FineMath-Llama-3B-Q5_K_S.gguf Q5_K_S 2.27GB false High quality, recommended.
FineMath-Llama-3B-Q4_K_L.gguf Q4_K_L 2.11GB false Uses Q8_0 for embed and output weights. Good quality, recommended.
FineMath-Llama-3B-Q4_1.gguf Q4_1 2.09GB false Legacy format, similar performance to Q4_K_S but with improved tokens/watt on Apple silicon.
FineMath-Llama-3B-Q4_K_M.gguf Q4_K_M 2.02GB false Good quality, default size for most use cases, recommended.
FineMath-Llama-3B-Q4_K_S.gguf Q4_K_S 1.93GB false Slightly lower quality with more space savings, recommended.
FineMath-Llama-3B-Q4_0.gguf Q4_0 1.92GB false Legacy format, offers online repacking for ARM and AVX CPU inference.
FineMath-Llama-3B-IQ4_NL.gguf IQ4_NL 1.92GB false Similar to IQ4_XS, but slightly larger. Offers online repacking for ARM CPU inference.
FineMath-Llama-3B-Q3_K_XL.gguf Q3_K_XL 1.91GB false Uses Q8_0 for embed and output weights. Lower quality but usable, good for low RAM availability.
FineMath-Llama-3B-IQ4_XS.gguf IQ4_XS 1.83GB false Decent quality, smaller than Q4_K_S with similar performance, recommended.
FineMath-Llama-3B-Q3_K_L.gguf Q3_K_L 1.82GB false Lower quality but usable, good for low RAM availability.
FineMath-Llama-3B-Q3_K_M.gguf Q3_K_M 1.69GB false Low quality.
FineMath-Llama-3B-IQ3_M.gguf IQ3_M 1.60GB false Medium-low quality, new method with decent performance comparable to Q3_K_M.
FineMath-Llama-3B-Q3_K_S.gguf Q3_K_S 1.54GB false Low quality, not recommended.
FineMath-Llama-3B-IQ3_XS.gguf IQ3_XS 1.48GB false Lower quality, new method with decent performance, slightly better than Q3_K_S.
FineMath-Llama-3B-Q2_K_L.gguf Q2_K_L 1.46GB false Uses Q8_0 for embed and output weights. Very low quality but surprisingly usable.
FineMath-Llama-3B-Q2_K.gguf Q2_K 1.36GB false Very low quality but surprisingly usable.
FineMath-Llama-3B-IQ2_M.gguf IQ2_M 1.23GB 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.

Downloading using huggingface-cli

ARM/AVX information

Previously, you would download Q4_0_4_4/4_8/8_8, and these would have their weights interleaved in memory in order to improve performance on ARM and AVX machines by loading up more data in one pass.

Now, however, there is something called "online repacking" for weights. details in this PR. If you use Q4_0 and your hardware would benefit from repacking weights, it will do it automatically on the fly.

As of llama.cpp build b4282 you will not be able to run the Q4_0_X_X files and will instead need to use Q4_0.

Additionally, if you want to get slightly better quality for , you can use IQ4_NL thanks to this PR which will also repack the weights for ARM, though only the 4_4 for now. The loading time may be slower but it will result in an overall speed incrase.

Which file should I choose?

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
1,461
GGUF
Hardware compatibility
Log In to add your hardware

2-bit

3-bit

4-bit

5-bit

6-bit

8-bit

16-bit

32-bit

Model tree for bartowski/FineMath-Llama-3B-GGUF

Quantized
(7)
this model

Dataset used to train bartowski/FineMath-Llama-3B-GGUF