Our new X account is live! Follow @wizwand_team for updates
WorkDL logo mark

LLM Decoding on Llama 70B 3.1

3,119.55Throughput

DeepFusionKernel

-121.09720.231,561.552,402.87Feb 12, 2026
Updated 4d ago

Evaluation Results

MethodLinks
2026.02
3,119.553.4
2026.02
3,016.4-
2026.02
1,635.945.5
2026.02
1,551.33-
2026.02
914.624.2
2026.02
878.08-
2026.02
493.6-
2026.02
478.66-
2026.02
470.634.2
2026.02
462.01-
2026.02
453.75-
2026.02
452.6-
2026.02
451.64-
2026.02
448.93-
2026.02
416.19-
2026.02
396.7-
2026.02
390.94-
2026.02
237.053.6
2026.02
228.9-
2026.02
140.15-
2026.02
138.16-
2026.02
135.76-
2026.02
132.66-
2026.02
131.75-
2026.02
129.81-
2026.02
126.81-
2026.02
126.1-
2026.02
125.1-
2026.02
121.1313.2
2026.02
119.3-
2026.02
118.56-
2026.02
107.05-
2026.02
62.223.4
2026.02
60.2-
2026.02
52.37-
2026.02
37.57-
2026.02
37.5-
2026.02
36.79-
2026.02
35.68-
2026.02
35.53-
2026.02
34.97-
2026.02
34.3-
2026.02
33.64-
2026.02
33.47-
2026.02
32.93-
2026.02
18.11-
2026.02
8.39-
2026.02
3.55-