unparallelled commited on
Commit
a009104
·
verified ·
1 Parent(s): ba6a188

Upload evals/2026-09-17/RULER_t64_table.txt with huggingface_hub

Browse files
evals/2026-09-17/RULER_t64_table.txt ADDED
@@ -0,0 +1,26 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ RULER coverage -- t64 (680M, 64k @1200 steps), N=100, virtual ~1,000 at every length
2
+ task cat 4096 8192 16360
3
+ ---------------------------------------------------------
4
+ niah_single retrieval 87.0 91.0 92.0
5
+ niah_single_1 retrieval 100.0 99.0 98.0
6
+ niah_single_3 retrieval 28.0 30.0 23.0
7
+ niah_multikey retrieval 43.0 53.0 44.0
8
+ niah_multikey_2 retrieval 27.0 25.0 35.0
9
+ niah_multikey_3 retrieval 10.0 4.0 13.0
10
+ niah_multivalue retrieval 0.0 0.0 0.0
11
+ niah_multiquery retrieval 0.0 0.0 0.0
12
+ vt multi-hop 3.0 3.0 2.0
13
+ cwe aggregation 0.0 0.0 0.0
14
+ fwe aggregation 0.0 0.0 0.0
15
+ qa_1 QA 3.0 7.0 8.0
16
+ qa_2 QA 16.0 12.0 17.0
17
+ ---------------------------------------------------------
18
+ mean over 13 tasks @ 4096: 24.4
19
+ mean over 13 tasks @ 8192: 24.9
20
+ mean over 13 tasks @ 16360: 25.5
21
+
22
+ by category @ 16,360:
23
+ retrieval 38.1 (n=8)
24
+ multi-hop 2.0 (n=1)
25
+ aggregation 0.0 (n=2)
26
+ QA 12.5 (n=2)