MetaIX commited on
Commit
dbc7df4
1 Parent(s): 09dc0f0

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +13 -1
README.md CHANGED
@@ -8,10 +8,22 @@ This was made using Chansung's GPT4-Alpaca Lora: https://huggingface.co/chansung
8
 
9
  <p><strong><font size="5">Benchmarks</font></strong></p>
10
 
 
 
11
  <strong>Wikitext2</strong>: 4.481280326843262
12
 
13
  <strong>Ptb-New</strong>: 8.539161682128906
14
 
15
  <strong>C4-New</strong>: 6.451964855194092
16
 
17
- <strong>Note</strong>: This version does not use <i>--groupsize 128</i>, therefore evaluations are minimally higher. However, this version allows fitting the whole model at full context using only 24GB VRAM.
 
 
 
 
 
 
 
 
 
 
 
8
 
9
  <p><strong><font size="5">Benchmarks</font></strong></p>
10
 
11
+ <p><strong><font size="4">--true-sequential --act-order</font></strong></p>
12
+
13
  <strong>Wikitext2</strong>: 4.481280326843262
14
 
15
  <strong>Ptb-New</strong>: 8.539161682128906
16
 
17
  <strong>C4-New</strong>: 6.451964855194092
18
 
19
+ <strong>Note</strong>: This version does not use <i>--groupsize 128</i>, therefore evaluations are minimally higher. However, this version allows fitting the whole model at full context using only 24GB VRAM.
20
+
21
+ <p><strong><font size="4">--true-sequential --groupsize 128</font></strong></p>
22
+
23
+ <strong>Wikitext2</strong>: 4.285132884979248
24
+
25
+ <strong>Ptb-New</strong>: 8.34856128692627
26
+
27
+ <strong>C4-New</strong>: 6.292652130126953
28
+
29
+ <strong>Note</strong>: This version uses <i>--groupsize 128</i>, resulting in better evaluations. However, it consumes more VRAM.