eachadea commited on
Commit
5db752d
1 Parent(s): a5a45bd

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +6 -1
README.md CHANGED
@@ -9,8 +9,13 @@ inference: true
9
  - Based on version 1.1
10
  - Used PR "More accurate Q4_0 and Q4_1 quantizations #896" (should be closer in quality to unquantized)
11
  - Uncensored variant is available, but it's based on version 1.0
 
 
 
 
 
 
12
  - 13B version of this can be found here: https://huggingface.co/eachadea/ggml-vicuna-13b-1.1
13
- - Choosing between q4_0 and q4_1, the logic of higher number \= better doesn't apply. If you're confused, stick with q4_0.
14
  <br>
15
  <br>
16
 
 
9
  - Based on version 1.1
10
  - Used PR "More accurate Q4_0 and Q4_1 quantizations #896" (should be closer in quality to unquantized)
11
  - Uncensored variant is available, but it's based on version 1.0
12
+ - For q4_2, "Q4_2 ARM #1046" was used. Will update regularly if new changes are made.
13
+ - **Choosing between q4_0, q4_1, and q4_2:**
14
+ - 4_0 is the fastest. The quality is the poorest.
15
+ - 4_1 is a lot slower. The quality is noticeably better.
16
+ - 4_2 is almost as fast as 4_0 and about as good as 4_1 **on Apple Silicon**. On Intel/AMD it's hardly better or faster than 4_1.
17
+
18
  - 13B version of this can be found here: https://huggingface.co/eachadea/ggml-vicuna-13b-1.1
 
19
  <br>
20
  <br>
21