bartowski
/

Gemma-2-9B-It-SPPO-Iter3-GGUF

@@ -1,29 +1,25 @@
 ---
-license: apache-2.0
-datasets:
-- openbmb/UltraFeedback
-language:
-- en
-pipeline_tag: text-generation
 quantized_by: bartowski
 ---
 ## Llamacpp imatrix Quantizations of Gemma-2-9B-It-SPPO-Iter3
-Using <a href="https://github.com/ggerganov/llama.cpp/">llama.cpp</a> release <a href="https://github.com/ggerganov/llama.cpp/releases/tag/b3278">b3278</a> for quantization.
 Original model: https://huggingface.co/UCLA-AGI/Gemma-2-9B-It-SPPO-Iter3
 All quants made using imatrix option with dataset from [here](https://gist.github.com/bartowski1182/eb213dccb3571f863da82e99418f81e8)
-Experimental quants are made with `--output-tensor-type f16 --token-embedding-type f16` per [ZeroWw](https://huggingface.co/ZeroWw)'s suggestion, please provide any feedback on quality differences you spot.
 ## Prompt format
 ```
-<start_of_turn>user
 {prompt}<end_of_turn>
 <start_of_turn>model
 ```
@@ -31,30 +27,34 @@ Note that this model does not support a System prompt.
 ## Download a file (not the whole branch) from below:
-| Filename | Quant type | File Size | Description |
-| -------- | ---------- | --------- | ----------- |
-| [Gemma-2-9B-It-SPPO-Iter3-Q8_0_L.gguf](https://huggingface.co/bartowski/Gemma-2-9B-It-SPPO-Iter3-GGUF/blob/main/Gemma-2-9B-It-SPPO-Iter3-Q8_1.gguf) | Q8_0_L | 10.68GB | *Experimental*, uses f16 for embed and output weights. Please provide any feedback of differences. Extremely high quality, generally unneeded but max available quant. |
-| [Gemma-2-9B-It-SPPO-Iter3-Q8_0.gguf](https://huggingface.co/bartowski/Gemma-2-9B-It-SPPO-Iter3-GGUF/blob/main/Gemma-2-9B-It-SPPO-Iter3-Q8_0.gguf) | Q8_0 | 9.82GB | Extremely high quality, generally unneeded but max available quant. |
-| [Gemma-2-9B-It-SPPO-Iter3-Q6_K_L.gguf](https://huggingface.co/bartowski/Gemma-2-9B-It-SPPO-Iter3-GGUF/blob/main/Gemma-2-9B-It-SPPO-Iter3-Q6_K_L.gguf) | Q6_K_L | 8.67GB | *Experimental*, uses f16 for embed and output weights. Please provide any feedback of differences. Very high quality, near perfect, *recommended*. |
-| [Gemma-2-9B-It-SPPO-Iter3-Q6_K.gguf](https://huggingface.co/bartowski/Gemma-2-9B-It-SPPO-Iter3-GGUF/blob/main/Gemma-2-9B-It-SPPO-Iter3-Q6_K.gguf) | Q6_K | 7.58GB | Very high quality, near perfect, *recommended*. |
-| [Gemma-2-9B-It-SPPO-Iter3-Q5_K_L.gguf](https://huggingface.co/bartowski/Gemma-2-9B-It-SPPO-Iter3-GGUF/blob/main/Gemma-2-9B-It-SPPO-Iter3-Q5_K_L.gguf) | Q5_K_L | 7.72GB | *Experimental*, uses f16 for embed and output weights. Please provide any feedback of differences. High quality, *recommended*. |
-| [Gemma-2-9B-It-SPPO-Iter3-Q5_K_M.gguf](https://huggingface.co/bartowski/Gemma-2-9B-It-SPPO-Iter3-GGUF/blob/main/Gemma-2-9B-It-SPPO-Iter3-Q5_K_M.gguf) | Q5_K_M | 6.64GB | High quality, *recommended*. |
-| [Gemma-2-9B-It-SPPO-Iter3-Q5_K_S.gguf](https://huggingface.co/bartowski/Gemma-2-9B-It-SPPO-Iter3-GGUF/blob/main/Gemma-2-9B-It-SPPO-Iter3-Q5_K_S.gguf) | Q5_K_S | 6.48GB | High quality, *recommended*. |
-| [Gemma-2-9B-It-SPPO-Iter3-Q4_K_L.gguf](https://huggingface.co/bartowski/Gemma-2-9B-It-SPPO-Iter3-GGUF/blob/main/Gemma-2-9B-It-SPPO-Iter3-Q4_K_L.gguf) | Q4_K_L | 6.84GB | *Experimental*, uses f16 for embed and output weights. Please provide any feedback of differences. Good quality, uses about 4.83 bits per weight, *recommended*. |
-| [Gemma-2-9B-It-SPPO-Iter3-Q4_K_M.gguf](https://huggingface.co/bartowski/Gemma-2-9B-It-SPPO-Iter3-GGUF/blob/main/Gemma-2-9B-It-SPPO-Iter3-Q4_K_M.gguf) | Q4_K_M | 5.76GB | Good quality, uses about 4.83 bits per weight, *recommended*. |
-| [Gemma-2-9B-It-SPPO-Iter3-Q4_K_S.gguf](https://huggingface.co/bartowski/Gemma-2-9B-It-SPPO-Iter3-GGUF/blob/main/Gemma-2-9B-It-SPPO-Iter3-Q4_K_S.gguf) | Q4_K_S | 5.47GB | Slightly lower quality with more space savings, *recommended*. |
-| [Gemma-2-9B-It-SPPO-Iter3-IQ4_XS.gguf](https://huggingface.co/bartowski/Gemma-2-9B-It-SPPO-Iter3-GGUF/blob/main/Gemma-2-9B-It-SPPO-Iter3-IQ4_XS.gguf) | IQ4_XS | 5.18GB | Decent quality, smaller than Q4_K_S with similar performance, *recommended*. |
-| [Gemma-2-9B-It-SPPO-Iter3-Q3_K_XL.gguf](https://huggingface.co/bartowski/Gemma-2-9B-It-SPPO-Iter3-GGUF/blob/main/Gemma-2-9B-It-SPPO-Iter3-Q3_K_XL.gguf) | Q3_K_XL | 6.21GB | *Experimental*, uses f16 for embed and output weights. Please provide any feedback of differences. Lower quality but usable, good for low RAM availability. |
-| [Gemma-2-9B-It-SPPO-Iter3-Q3_K_L.gguf](https://huggingface.co/bartowski/Gemma-2-9B-It-SPPO-Iter3-GGUF/blob/main/Gemma-2-9B-It-SPPO-Iter3-Q3_K_L.gguf) | Q3_K_L | 5.13GB | Lower quality but usable, good for low RAM availability. |
-| [Gemma-2-9B-It-SPPO-Iter3-Q3_K_M.gguf](https://huggingface.co/bartowski/Gemma-2-9B-It-SPPO-Iter3-GGUF/blob/main/Gemma-2-9B-It-SPPO-Iter3-Q3_K_M.gguf) | Q3_K_M | 4.76GB | Even lower quality. |
-| [Gemma-2-9B-It-SPPO-Iter3-IQ3_M.gguf](https://huggingface.co/bartowski/Gemma-2-9B-It-SPPO-Iter3-GGUF/blob/main/Gemma-2-9B-It-SPPO-Iter3-IQ3_M.gguf) | IQ3_M | 4.49GB | Medium-low quality, new method with decent performance comparable to Q3_K_M. |
-| [Gemma-2-9B-It-SPPO-Iter3-Q3_K_S.gguf](https://huggingface.co/bartowski/Gemma-2-9B-It-SPPO-Iter3-GGUF/blob/main/Gemma-2-9B-It-SPPO-Iter3-Q3_K_S.gguf) | Q3_K_S | 4.33GB | Low quality, not recommended. |
-| [Gemma-2-9B-It-SPPO-Iter3-IQ3_XS.gguf](https://huggingface.co/bartowski/Gemma-2-9B-It-SPPO-Iter3-GGUF/blob/main/Gemma-2-9B-It-SPPO-Iter3-IQ3_XS.gguf) | IQ3_XS | 4.14GB | Lower quality, new method with decent performance, slightly better than Q3_K_S. |
-| [Gemma-2-9B-It-SPPO-Iter3-IQ3_XXS.gguf](https://huggingface.co/bartowski/Gemma-2-9B-It-SPPO-Iter3-GGUF/blob/main/Gemma-2-9B-It-SPPO-Iter3-IQ3_XXS.gguf) | IQ3_XXS | 3.79GB | Lower quality, new method with decent performance, comparable to Q3 quants. |
-| [Gemma-2-9B-It-SPPO-Iter3-Q2_K.gguf](https://huggingface.co/bartowski/Gemma-2-9B-It-SPPO-Iter3-GGUF/blob/main/Gemma-2-9B-It-SPPO-Iter3-Q2_K.gguf) | Q2_K | 3.80GB | Very low quality but surprisingly usable. |
-| [Gemma-2-9B-It-SPPO-Iter3-IQ2_M.gguf](https://huggingface.co/bartowski/Gemma-2-9B-It-SPPO-Iter3-GGUF/blob/main/Gemma-2-9B-It-SPPO-Iter3-IQ2_M.gguf) | IQ2_M | 3.43GB | Very low quality, uses SOTA techniques to also be surprisingly usable. |
-| [Gemma-2-9B-It-SPPO-Iter3-IQ2_S.gguf](https://huggingface.co/bartowski/Gemma-2-9B-It-SPPO-Iter3-GGUF/blob/main/Gemma-2-9B-It-SPPO-Iter3-IQ2_S.gguf) | IQ2_S | 3.21GB | Very low quality, uses SOTA techniques to be usable. |
-| [Gemma-2-9B-It-SPPO-Iter3-IQ2_XS.gguf](https://huggingface.co/bartowski/Gemma-2-9B-It-SPPO-Iter3-GGUF/blob/main/Gemma-2-9B-It-SPPO-Iter3-IQ2_XS.gguf) | IQ2_XS | 3.06GB | Very low quality, uses SOTA techniques to be usable. |
 ## Downloading using huggingface-cli
@@ -103,3 +103,4 @@ These I-quants can also be used on CPU and Apple Metal, but will be slower than
 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

 ---
 quantized_by: bartowski
+pipeline_tag: text-generation
 ---
 ## Llamacpp imatrix Quantizations of Gemma-2-9B-It-SPPO-Iter3
+Using <a href="https://github.com/ggerganov/llama.cpp/">llama.cpp</a> release <a href="https://github.com/ggerganov/llama.cpp/releases/tag/b3389">b3389</a> for quantization.
 Original model: https://huggingface.co/UCLA-AGI/Gemma-2-9B-It-SPPO-Iter3
 All quants made using imatrix option with dataset from [here](https://gist.github.com/bartowski1182/eb213dccb3571f863da82e99418f81e8)
 ## Prompt format
 ```
+<bos><start_of_turn>user
 {prompt}<end_of_turn>
 <start_of_turn>model
+<end_of_turn>
+<start_of_turn>model
 ```
 ## Download a file (not the whole branch) from below:
+| Filename | Quant type | File Size | Split | Description |
+| -------- | ---------- | --------- | ----- | ----------- |
+| [Gemma-2-9B-It-SPPO-Iter3-f32.gguf](https://huggingface.co/bartowski/Gemma-2-9B-It-SPPO-Iter3-GGUF/blob/main/Gemma-2-9B-It-SPPO-Iter3-f32.gguf) | f32 | 36.97GB | false | Full F32 weights. |
+| [Gemma-2-9B-It-SPPO-Iter3-Q8_0.gguf](https://huggingface.co/bartowski/Gemma-2-9B-It-SPPO-Iter3-GGUF/blob/main/Gemma-2-9B-It-SPPO-Iter3-Q8_0.gguf) | Q8_0 | 9.83GB | false | Extremely high quality, generally unneeded but max available quant. |
+| [Gemma-2-9B-It-SPPO-Iter3-Q6_K_L.gguf](https://huggingface.co/bartowski/Gemma-2-9B-It-SPPO-Iter3-GGUF/blob/main/Gemma-2-9B-It-SPPO-Iter3-Q6_K_L.gguf) | Q6_K_L | 7.81GB | false | Uses Q8_0 for embed and output weights. Very high quality, near perfect, *recommended*. |
+| [Gemma-2-9B-It-SPPO-Iter3-Q6_K.gguf](https://huggingface.co/bartowski/Gemma-2-9B-It-SPPO-Iter3-GGUF/blob/main/Gemma-2-9B-It-SPPO-Iter3-Q6_K.gguf) | Q6_K | 7.59GB | false | Very high quality, near perfect, *recommended*. |
+| [Gemma-2-9B-It-SPPO-Iter3-Q5_K_L.gguf](https://huggingface.co/bartowski/Gemma-2-9B-It-SPPO-Iter3-GGUF/blob/main/Gemma-2-9B-It-SPPO-Iter3-Q5_K_L.gguf) | Q5_K_L | 6.87GB | false | Uses Q8_0 for embed and output weights. High quality, *recommended*. |
+| [Gemma-2-9B-It-SPPO-Iter3-Q5_K_M.gguf](https://huggingface.co/bartowski/Gemma-2-9B-It-SPPO-Iter3-GGUF/blob/main/Gemma-2-9B-It-SPPO-Iter3-Q5_K_M.gguf) | Q5_K_M | 6.65GB | false | High quality, *recommended*. |
+| [Gemma-2-9B-It-SPPO-Iter3-Q5_K_S.gguf](https://huggingface.co/bartowski/Gemma-2-9B-It-SPPO-Iter3-GGUF/blob/main/Gemma-2-9B-It-SPPO-Iter3-Q5_K_S.gguf) | Q5_K_S | 6.48GB | false | High quality, *recommended*. |
+| [Gemma-2-9B-It-SPPO-Iter3-Q4_K_L.gguf](https://huggingface.co/bartowski/Gemma-2-9B-It-SPPO-Iter3-GGUF/blob/main/Gemma-2-9B-It-SPPO-Iter3-Q4_K_L.gguf) | Q4_K_L | 5.98GB | false | Uses Q8_0 for embed and output weights. Good quality, *recommended*. |
+| [Gemma-2-9B-It-SPPO-Iter3-Q4_K_M.gguf](https://huggingface.co/bartowski/Gemma-2-9B-It-SPPO-Iter3-GGUF/blob/main/Gemma-2-9B-It-SPPO-Iter3-Q4_K_M.gguf) | Q4_K_M | 5.76GB | false | Good quality, default size for must use cases, *recommended*. |
+| [Gemma-2-9B-It-SPPO-Iter3-Q4_K_S.gguf](https://huggingface.co/bartowski/Gemma-2-9B-It-SPPO-Iter3-GGUF/blob/main/Gemma-2-9B-It-SPPO-Iter3-Q4_K_S.gguf) | Q4_K_S | 5.48GB | false | Slightly lower quality with more space savings, *recommended*. |
+| [Gemma-2-9B-It-SPPO-Iter3-IQ4_XS.gguf](https://huggingface.co/bartowski/Gemma-2-9B-It-SPPO-Iter3-GGUF/blob/main/Gemma-2-9B-It-SPPO-Iter3-IQ4_XS.gguf) | IQ4_XS | 5.18GB | false | Decent quality, smaller than Q4_K_S with similar performance, *recommended*. |
+| [Gemma-2-9B-It-SPPO-Iter3-Q3_K_L.gguf](https://huggingface.co/bartowski/Gemma-2-9B-It-SPPO-Iter3-GGUF/blob/main/Gemma-2-9B-It-SPPO-Iter3-Q3_K_L.gguf) | Q3_K_L | 5.13GB | false | Lower quality but usable, good for low RAM availability. |
+| [Gemma-2-9B-It-SPPO-Iter3-Q3_K_M.gguf](https://huggingface.co/bartowski/Gemma-2-9B-It-SPPO-Iter3-GGUF/blob/main/Gemma-2-9B-It-SPPO-Iter3-Q3_K_M.gguf) | Q3_K_M | 4.76GB | false | Low quality. |
+| [Gemma-2-9B-It-SPPO-Iter3-IQ3_M.gguf](https://huggingface.co/bartowski/Gemma-2-9B-It-SPPO-Iter3-GGUF/blob/main/Gemma-2-9B-It-SPPO-Iter3-IQ3_M.gguf) | IQ3_M | 4.49GB | false | Medium-low quality, new method with decent performance comparable to Q3_K_M. |
+| [Gemma-2-9B-It-SPPO-Iter3-Q3_K_S.gguf](https://huggingface.co/bartowski/Gemma-2-9B-It-SPPO-Iter3-GGUF/blob/main/Gemma-2-9B-It-SPPO-Iter3-Q3_K_S.gguf) | Q3_K_S | 4.34GB | false | Low quality, not recommended. |
+| [Gemma-2-9B-It-SPPO-Iter3-IQ3_XS.gguf](https://huggingface.co/bartowski/Gemma-2-9B-It-SPPO-Iter3-GGUF/blob/main/Gemma-2-9B-It-SPPO-Iter3-IQ3_XS.gguf) | IQ3_XS | 4.14GB | false | Lower quality, new method with decent performance, slightly better than Q3_K_S. |
+| [Gemma-2-9B-It-SPPO-Iter3-Q2_K_L.gguf](https://huggingface.co/bartowski/Gemma-2-9B-It-SPPO-Iter3-GGUF/blob/main/Gemma-2-9B-It-SPPO-Iter3-Q2_K_L.gguf) | Q2_K_L | 4.03GB | false | Uses Q8_0 for embed and output weights. Very low quality but surprisingly usable. |
+| [Gemma-2-9B-It-SPPO-Iter3-Q2_K.gguf](https://huggingface.co/bartowski/Gemma-2-9B-It-SPPO-Iter3-GGUF/blob/main/Gemma-2-9B-It-SPPO-Iter3-Q2_K.gguf) | Q2_K | 3.81GB | false | Very low quality but surprisingly usable. |
+| [Gemma-2-9B-It-SPPO-Iter3-IQ3_XXS.gguf](https://huggingface.co/bartowski/Gemma-2-9B-It-SPPO-Iter3-GGUF/blob/main/Gemma-2-9B-It-SPPO-Iter3-IQ3_XXS.gguf) | IQ3_XXS | 3.80GB | false | Lower quality, new method with decent performance, comparable to Q3 quants. |
+| [Gemma-2-9B-It-SPPO-Iter3-IQ2_M.gguf](https://huggingface.co/bartowski/Gemma-2-9B-It-SPPO-Iter3-GGUF/blob/main/Gemma-2-9B-It-SPPO-Iter3-IQ2_M.gguf) | IQ2_M | 3.43GB | false | Relatively low quality, uses SOTA techniques to be surprisingly usable. |
+## 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
 ## Downloading using huggingface-cli
 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