Share your thoughts, 1 month free Claude Pro on us
See more
Home
/
Benchmarks
Long-context Retrieval on RULER 8k context NIAH Single 1
Loading...
100
Accuracy
BF16
-3.168
23.616
50.4
77.184
May 18, 2026
Accuracy
Score
Relative Change (%)
Average Gate Density
Updated 15d ago
Evaluation Results
Method
Method
Links
Accuracy
Score
Relative Change (%)
Average Gate Density
BF16
Backbone=Qwen3-4B-Thin...
2026.05
100
-
-
-
OSCAR
Backbone=Qwen3-4B-Thin...
2026.05
100
-
-
-
BF16
Backbone=GLM-4.7-FP8†
2026.05
100
-
-
-
OSCAR
Backbone=GLM-4.7-FP8†
2026.05
100
-
-
-
QuaRot-INT2
Backbone=GLM-4.7-FP8†
2026.05
99.7
-
-
-
BF16
Backbone=Qwen3-8B
2026.05
99.6
-
-
-
OSCAR
Backbone=Qwen3-8B
2026.05
97.8
-
-
-
QuaRot-INT2
Backbone=Qwen3-8B
2026.05
80.6
-
-
-
QuaRot-INT2
Backbone=Qwen3-4B-Thin...
2026.05
0.8
-
-
-
Vanilla
Attention Mechanism=Fu...
2026.05
-
1
-
-
Self-Pruned KV
Attention Mechanism=Se...
2026.05
-
1
0
5.6
Feedback
Search any
task
Search any
task