Upload README.md with huggingface_hub

README.md

@@ -1,15 +1,11 @@
 ---
-base_model: IlyaGusev/gemma-2-9b-it-abliterated
-language:
-- en
-license: gemma
-pipeline_tag: text-generation
 quantized_by: bartowski
+pipeline_tag: text-generation
 ---
 
 ## Llamacpp imatrix Quantizations of gemma-2-9b-it-abliterated
 
-Using <a href="https://github.com/ggerganov/llama.cpp/">llama.cpp</a> release <a href="https://github.com/ggerganov/llama.cpp/releases/tag/b3427">b3427</a> for quantization.
+Using <a href="https://github.com/ggerganov/llama.cpp/">llama.cpp</a> release <a href="https://github.com/ggerganov/llama.cpp/releases/tag/b3878">b3878</a> for quantization.
 
 Original model: https://huggingface.co/IlyaGusev/gemma-2-9b-it-abliterated
 
@@ -27,7 +23,6 @@ Run them in [LM Studio](https://lmstudio.ai/)
 <start_of_turn>model
 <end_of_turn>
 <start_of_turn>model
-
 ```
 
 ## Download a file (not the whole branch) from below:
@@ -35,6 +30,7 @@ Run them in [LM Studio](https://lmstudio.ai/)
 | Filename | Quant type | File Size | Split | Description |
 | -------- | ---------- | --------- | ----- | ----------- |
 | [gemma-2-9b-it-abliterated-f32.gguf](https://huggingface.co/bartowski/gemma-2-9b-it-abliterated-GGUF/blob/main/gemma-2-9b-it-abliterated-f32.gguf) | f32 | 36.97GB | false | Full F32 weights. |
+| [gemma-2-9b-it-abliterated-f32.gguf](https://huggingface.co/bartowski/gemma-2-9b-it-abliterated-GGUF/blob/main/gemma-2-9b-it-abliterated-f32.gguf) | f32 | 36.97GB | false | Full F32 weights. |
 | [gemma-2-9b-it-abliterated-Q8_0.gguf](https://huggingface.co/bartowski/gemma-2-9b-it-abliterated-GGUF/blob/main/gemma-2-9b-it-abliterated-Q8_0.gguf) | Q8_0 | 9.83GB | false | Extremely high quality, generally unneeded but max available quant. |
 | [gemma-2-9b-it-abliterated-Q6_K_L.gguf](https://huggingface.co/bartowski/gemma-2-9b-it-abliterated-GGUF/blob/main/gemma-2-9b-it-abliterated-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-abliterated-Q6_K.gguf](https://huggingface.co/bartowski/gemma-2-9b-it-abliterated-GGUF/blob/main/gemma-2-9b-it-abliterated-Q6_K.gguf) | Q6_K | 7.59GB | false | Very high quality, near perfect, *recommended*. |
@@ -44,6 +40,9 @@ Run them in [LM Studio](https://lmstudio.ai/)
 | [gemma-2-9b-it-abliterated-Q4_K_L.gguf](https://huggingface.co/bartowski/gemma-2-9b-it-abliterated-GGUF/blob/main/gemma-2-9b-it-abliterated-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-abliterated-Q4_K_M.gguf](https://huggingface.co/bartowski/gemma-2-9b-it-abliterated-GGUF/blob/main/gemma-2-9b-it-abliterated-Q4_K_M.gguf) | Q4_K_M | 5.76GB | false | Good quality, default size for must use cases, *recommended*. |
 | [gemma-2-9b-it-abliterated-Q4_K_S.gguf](https://huggingface.co/bartowski/gemma-2-9b-it-abliterated-GGUF/blob/main/gemma-2-9b-it-abliterated-Q4_K_S.gguf) | Q4_K_S | 5.48GB | false | Slightly lower quality with more space savings, *recommended*. |
+| [gemma-2-9b-it-abliterated-Q4_0_8_8.gguf](https://huggingface.co/bartowski/gemma-2-9b-it-abliterated-GGUF/blob/main/gemma-2-9b-it-abliterated-Q4_0_8_8.gguf) | Q4_0_8_8 | 5.44GB | false | Optimized for ARM inference. Requires 'sve' support (see link below). |
+| [gemma-2-9b-it-abliterated-Q4_0_4_8.gguf](https://huggingface.co/bartowski/gemma-2-9b-it-abliterated-GGUF/blob/main/gemma-2-9b-it-abliterated-Q4_0_4_8.gguf) | Q4_0_4_8 | 5.44GB | false | Optimized for ARM inference. Requires 'i8mm' support (see link below). |
+| [gemma-2-9b-it-abliterated-Q4_0_4_4.gguf](https://huggingface.co/bartowski/gemma-2-9b-it-abliterated-GGUF/blob/main/gemma-2-9b-it-abliterated-Q4_0_4_4.gguf) | Q4_0_4_4 | 5.44GB | false | Optimized for ARM inference. Should work well on all ARM chips, pick this if you're unsure. |
 | [gemma-2-9b-it-abliterated-Q3_K_XL.gguf](https://huggingface.co/bartowski/gemma-2-9b-it-abliterated-GGUF/blob/main/gemma-2-9b-it-abliterated-Q3_K_XL.gguf) | Q3_K_XL | 5.35GB | false | Uses Q8_0 for embed and output weights. Lower quality but usable, good for low RAM availability. |
 | [gemma-2-9b-it-abliterated-IQ4_XS.gguf](https://huggingface.co/bartowski/gemma-2-9b-it-abliterated-GGUF/blob/main/gemma-2-9b-it-abliterated-IQ4_XS.gguf) | IQ4_XS | 5.18GB | false | Decent quality, smaller than Q4_K_S with similar performance, *recommended*. |
 | [gemma-2-9b-it-abliterated-Q3_K_L.gguf](https://huggingface.co/bartowski/gemma-2-9b-it-abliterated-GGUF/blob/main/gemma-2-9b-it-abliterated-Q3_K_L.gguf) | Q3_K_L | 5.13GB | false | Lower quality but usable, good for low RAM availability. |
@@ -55,11 +54,13 @@ Run them in [LM Studio](https://lmstudio.ai/)
 | [gemma-2-9b-it-abliterated-Q2_K.gguf](https://huggingface.co/bartowski/gemma-2-9b-it-abliterated-GGUF/blob/main/gemma-2-9b-it-abliterated-Q2_K.gguf) | Q2_K | 3.81GB | false | Very low quality but surprisingly usable. |
 | [gemma-2-9b-it-abliterated-IQ2_M.gguf](https://huggingface.co/bartowski/gemma-2-9b-it-abliterated-GGUF/blob/main/gemma-2-9b-it-abliterated-IQ2_M.gguf) | IQ2_M | 3.43GB | false | Relatively low quality, uses SOTA techniques to be surprisingly usable. |
 
-## Credits
+## Embed/output weights
 
-Thank you kalomaze and Dampf for assistance in creating the imatrix calibration dataset
+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.
 
-Thank you ZeroWw for the inspiration to experiment with embed/output
+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
 
@@ -78,11 +79,19 @@ huggingface-cli download bartowski/gemma-2-9b-it-abliterated-GGUF --include "gem
 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/gemma-2-9b-it-abliterated-GGUF --include "gemma-2-9b-it-abliterated-Q8_0.gguf/*" --local-dir gemma-2-9b-it-abliterated-Q8_0
+huggingface-cli download bartowski/gemma-2-9b-it-abliterated-GGUF --include "gemma-2-9b-it-abliterated-Q8_0/*" --local-dir ./
 ```
 
 You can either specify a new local-dir (gemma-2-9b-it-abliterated-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](https://github.com/ggerganov/llama.cpp/pull/5780#pullrequestreview-21657544660)
+
+To check which one would work best for your ARM chip, you can check [AArch64 SoC features](https://gpages.juszkiewicz.com.pl/arm-socs-table/arm-socs.html) (thanks EloyOn!).
+
 ## Which file should I choose?
 
 A great write up with charts showing various performances is provided by Artefact2 [here](https://gist.github.com/Artefact2/b5f810600771265fc1e39442288e8ec9)
@@ -107,5 +116,10 @@ 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
+## 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