neopolita commited on
Commit
39b63d6
1 Parent(s): be64d4d

Upload README.md with huggingface_hub

Browse files
Files changed (1) hide show
  1. README.md +27 -0
README.md ADDED
@@ -0,0 +1,27 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ {}
3
+ ---
4
+ # GGUF quants for [**google/gemma-2-9b-it**](https://huggingface.co/google/gemma-2-9b-it) using [llama.cpp](https://github.com/ggerganov/llama.cpp)
5
+
6
+ **Terms of Use**: Please check the [**original model**](https://huggingface.co/google/gemma-2-9b-it)
7
+
8
+ <picture>
9
+ <img alt="cthulhu" src="https://huggingface.co/neopolita/common/resolve/main/profile.png">
10
+ </picture>
11
+
12
+ ## Quants
13
+
14
+ * `q2_k`: Uses Q4_K for the attention.vw and feed_forward.w2 tensors, Q2_K for the other tensors.
15
+ * `q3_k_s`: Uses Q3_K for all tensors
16
+ * `q3_k_m`: Uses Q4_K for the attention.wv, attention.wo, and feed_forward.w2 tensors, else Q3_K
17
+ * `q3_k_l`: Uses Q5_K for the attention.wv, attention.wo, and feed_forward.w2 tensors, else Q3_K
18
+ * `q4_0`: Original quant method, 4-bit.
19
+ * `q4_1`: Higher accuracy than q4_0 but not as high as q5_0. However has quicker inference than q5 models.
20
+ * `q4_k_s`: Uses Q4_K for all tensors
21
+ * `q4_k_m`: Uses Q6_K for half of the attention.wv and feed_forward.w2 tensors, else Q4_K
22
+ * `q5_0`: Higher accuracy, higher resource usage and slower inference.
23
+ * `q5_1`: Even higher accuracy, resource usage and slower inference.
24
+ * `q5_k_s`: Uses Q5_K for all tensors
25
+ * `q5_k_m`: Uses Q6_K for half of the attention.wv and feed_forward.w2 tensors, else Q5_K
26
+ * `q6_k`: Uses Q8_K for all tensors
27
+ * `q8_0`: Almost indistinguishable from float16. High resource use and slow. Not recommended for most users.