Gemma-4-26B-A4B-it-Uncensored (NVFP4)
Up to 1937 tok/s aggregate at c=64 — scales without the pre-fix concurrency crash
Aggregate throughput (tok/s) vs concurrent requests · DFlash on aeon-vllm-ultimate:latest · scales cleanly to c=64
0
532
1065
1598
2131
c=1
c=8
c=16
c=32
c=64
Coding
Math
Reasoning
Prose
Natural language
Extraction/JSON