Update README.md
Browse files
README.md
CHANGED
@@ -23,12 +23,29 @@ Master
|
|
23 |
Size : 3.52 GiB (3.76 BPW)
|
24 |
PPL 512 wikitext : 7.9263 +/- 0.04943
|
25 |
|
26 |
-
|
27 |
-
|
28 |
-
PR
|
29 |
Size : 3.49 GiB (3.73 BPW)
|
30 |
PPL 512 wikitext : 7.8704 +/- 0.04951
|
31 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
32 |
IQ4_XS
|
33 |
|
34 |
Master
|
@@ -39,7 +56,7 @@ PPL 512 wikitext : 7.5226 +/- 0.04820
|
|
39 |
|
40 |
IQ4_XSR
|
41 |
|
42 |
-
PR
|
43 |
Size : 4.16 GiB (4.45 BPW)
|
44 |
Arc-C 299
|
45 |
Arc-E 570
|
|
|
23 |
Size : 3.52 GiB (3.76 BPW)
|
24 |
PPL 512 wikitext : 7.9263 +/- 0.04943
|
25 |
|
26 |
+
PR (good)
|
|
|
|
|
27 |
Size : 3.49 GiB (3.73 BPW)
|
28 |
PPL 512 wikitext : 7.8704 +/- 0.04951
|
29 |
|
30 |
+
IQ3_XL
|
31 |
+
|
32 |
+
PR (good)
|
33 |
+
Size : 3.71 GiB (3.97 BPW)
|
34 |
+
PPL 512 wikitext : 7.7225 +/- 0.04946
|
35 |
+
|
36 |
+
IQ3_XXL
|
37 |
+
|
38 |
+
PR (good, the benefit seems meager but the token embeddings pushed form IQ3_S to IQ4_XS explains +0.05BPW of it,
|
39 |
+
and this tensor doesn't run in VRAM but in RAM)
|
40 |
+
Size : 3.83 GiB (4.09 BPW)
|
41 |
+
PPL 512 wikitext : 7.6720 +/- 0.04892
|
42 |
+
|
43 |
+
IQ3_XXL
|
44 |
+
|
45 |
+
PR (good)
|
46 |
+
Size : 3.97 GiB (4.24 BPW)
|
47 |
+
PPL 512 wikitext : 7.5920 +/- 0.04839
|
48 |
+
|
49 |
IQ4_XS
|
50 |
|
51 |
Master
|
|
|
56 |
|
57 |
IQ4_XSR
|
58 |
|
59 |
+
PR (good)
|
60 |
Size : 4.16 GiB (4.45 BPW)
|
61 |
Arc-C 299
|
62 |
Arc-E 570
|