Update README.md

README.md

@@ -13,10 +13,19 @@ base_model: google/gemma-2-9b-it
 
 # Quant Infos
 
-- f32 gguf is from the official kaggle repo
-- imatrix quants are running and will be uploaded one-by-one
-- you will need the gemma2 llama.cpp [PR](https://github.com/ggerganov/llama.cpp/pull/8156) applied to your llama.cpp
-- current quants are based on the f32 gguf provided by google directly, I will reconvert from the huggingface repo once the dust settles to get better gguf metadata
+- quants done with an importance matrix for improved quantization loss
+- Currently requantizing ggufs & imatrix from bf16 for "optimal" accuracy loss
+  - initial version was based on f32 gguf provided by google
+  - WIP new version should have better metadata as its quantized from scratch with llama.cpp
+- Wide coverage of different gguf quant types from Q\_8\_0 down to IQ1\_S
+  - experimental custom quant types
+    - `_L` with `--output-tensor-type f16 --token-embedding-type f16` (same as bartowski)
+    - `_XL` with `--output-tensor-type bf16 --token-embedding-type bf16` (same size as _L, in theory even higher numerical accuracy)
+- Quantized with [llama.cpp](https://github.com/ggerganov/llama.cpp) release [b3259](https://github.com/ggerganov/llama.cpp/releases/tag/b3259)
+- Imatrix generated with [this](https://gist.github.com/bartowski1182/eb213dccb3571f863da82e99418f81e8) multi-purpose dataset by [bartowski](https://huggingface.co/bartowski).
+  ```
+  ./imatrix -c 512 -m $model_name-bf16.gguf -f calibration_datav3.txt -o $model_name.imatrix
+  ```
 
 # Original Model Card