Llamacpp imatrix Quantizations of gemma-4-12B-it by google
Using llama.cpp release b9496 for quantization.
Original model: https://huggingface.co/google/gemma-4-12B-it
All quants made using imatrix option with dataset from here
Run them in your choice of tools:
Note: if it's a newly supported model, you may need to wait for an update from the developers.
What's new
Updated to latest model release from Google
Prompt format
<bos><|turn>system
{system_prompt}<turn|>
<|turn>user
{prompt}<turn|>
<|turn>model
<|channel>thought
<channel|>
Download a file (not the whole branch) from below:
| Filename | Quant type | File Size | Split | Description |
|---|---|---|---|---|
| gemma-4-12B-it-bf16.gguf | bf16 | 23.83GB | false | Full BF16 weights. |
| gemma-4-12B-it-Q8_0.gguf | Q8_0 | 12.67GB | false | Extremely high quality, generally unneeded but max available quant. |
| gemma-4-12B-it-Q6_K_L.gguf | Q6_K_L | 10.48GB | false | Uses Q8_0 for embed and output weights. Very high quality, near perfect, recommended. |
| gemma-4-12B-it-Q6_K.gguf | Q6_K | 10.24GB | false | Very high quality, near perfect, recommended. |
| gemma-4-12B-it-Q5_K_L.gguf | Q5_K_L | 9.02GB | false | Uses Q8_0 for embed and output weights. High quality, recommended. |
| gemma-4-12B-it-Q5_K_M.gguf | Q5_K_M | 8.77GB | false | High quality, recommended. |
| gemma-4-12B-it-Q5_K_S.gguf | Q5_K_S | 8.41GB | false | High quality, recommended. |
| gemma-4-12B-it-Q4_K_L.gguf | Q4_K_L | 7.91GB | false | Uses Q8_0 for embed and output weights. Good quality, recommended. |
| gemma-4-12B-it-Q4_1.gguf | Q4_1 | 7.76GB | false | Legacy format, similar performance to Q4_K_S but with improved tokens/watt on Apple silicon. |
| gemma-4-12B-it-Q4_K_M.gguf | Q4_K_M | 7.66GB | false | Good quality, default size for most use cases, recommended. |
| gemma-4-12B-it-Q4_K_S.gguf | Q4_K_S | 7.17GB | false | Slightly lower quality with more space savings, recommended. |
| gemma-4-12B-it-Q4_0.gguf | Q4_0 | 7.13GB | false | Legacy format, offers online repacking for ARM and AVX CPU inference. |
| gemma-4-12B-it-IQ4_NL.gguf | IQ4_NL | 7.11GB | false | Similar to IQ4_XS, but slightly larger. Offers online repacking for ARM CPU inference. |
| gemma-4-12B-it-Q3_K_XL.gguf | Q3_K_XL | 6.90GB | false | Uses Q8_0 for embed and output weights. Lower quality but usable, good for low RAM availability. |
| gemma-4-12B-it-IQ4_XS.gguf | IQ4_XS | 6.78GB | false | Decent quality, smaller than Q4_K_S with similar performance, recommended. |
| gemma-4-12B-it-Q3_K_L.gguf | Q3_K_L | 6.65GB | false | Lower quality but usable, good for low RAM availability. |
| gemma-4-12B-it-Q3_K_M.gguf | Q3_K_M | 6.30GB | false | Low quality. |
| gemma-4-12B-it-IQ3_M.gguf | IQ3_M | 5.97GB | false | Medium-low quality, new method with decent performance comparable to Q3_K_M. |
| gemma-4-12B-it-Q3_K_S.gguf | Q3_K_S | 5.72GB | false | Low quality, not recommended. |
| gemma-4-12B-it-IQ3_XS.gguf | IQ3_XS | 5.53GB | false | Lower quality, new method with decent performance, slightly better than Q3_K_S. |
| gemma-4-12B-it-Q2_K_L.gguf | Q2_K_L | 5.32GB | false | Uses Q8_0 for embed and output weights. Very low quality but surprisingly usable. |
| gemma-4-12B-it-IQ3_XXS.gguf | IQ3_XXS | 5.15GB | false | Lower quality, new method with decent performance, comparable to Q3 quants. |
| gemma-4-12B-it-Q2_K.gguf | Q2_K | 5.08GB | false | Very low quality but surprisingly usable. |
| gemma-4-12B-it-IQ2_M.gguf | IQ2_M | 4.94GB | false | Relatively low quality, uses SOTA techniques to be surprisingly usable. |
| gemma-4-12B-it-IQ2_S.gguf | IQ2_S | 4.71GB | false | Low quality, uses SOTA techniques to be 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.
Thank you to LM Studio for sponsoring my work.
Want to support my work? Visit my ko-fi page here: https://ko-fi.com/bartowski
- Downloads last month
- 54,886