Update README.md
Browse files
README.md
CHANGED
@@ -19,12 +19,10 @@ tags:
|
|
19 |
Fully rebuilt from master files, including full merge(s) to maintain full 32 bit precision right
|
20 |
up until it is compressed into GGUF files which results on a top to bottom upgrade.
|
21 |
|
22 |
-
The result is
|
23 |
|
24 |
<img src="dark-forest.jpg">
|
25 |
|
26 |
-
Better prose, better instruction following and stronger creative brainpower.
|
27 |
-
|
28 |
On average this means a q4km operates at Q6 levels and Q6 and Q8 exceeds original model full precision performance.
|
29 |
|
30 |
Perplexity drop (lower is better) is close to 10% (over 600 points for q4km) for all quants.
|
@@ -32,8 +30,6 @@ Perplexity drop (lower is better) is close to 10% (over 600 points for q4km) for
|
|
32 |
That means precision has been enhanced for all 20 billion parameters which affects "brain density" / "function",
|
33 |
instruction following and output quality.
|
34 |
|
35 |
-
Quants uploading now.
|
36 |
-
|
37 |
Imatrix quants to follow shortly.
|
38 |
|
39 |
For more details, including a list of enhancements see our other 32 bit
|
|
|
19 |
Fully rebuilt from master files, including full merge(s) to maintain full 32 bit precision right
|
20 |
up until it is compressed into GGUF files which results on a top to bottom upgrade.
|
21 |
|
22 |
+
The result is superior performance in instruction following, reasoning, depth, nuance and emotion.
|
23 |
|
24 |
<img src="dark-forest.jpg">
|
25 |
|
|
|
|
|
26 |
On average this means a q4km operates at Q6 levels and Q6 and Q8 exceeds original model full precision performance.
|
27 |
|
28 |
Perplexity drop (lower is better) is close to 10% (over 600 points for q4km) for all quants.
|
|
|
30 |
That means precision has been enhanced for all 20 billion parameters which affects "brain density" / "function",
|
31 |
instruction following and output quality.
|
32 |
|
|
|
|
|
33 |
Imatrix quants to follow shortly.
|
34 |
|
35 |
For more details, including a list of enhancements see our other 32 bit
|