List your company

Code Generation on HumanEval (TPS, Latency)

1,796.9TPS (Tokens/s)DFLARE
-70.68414.1725899.0251,383.8775Feb 12, 2026Mar 2, 2026Mar 20, 2026Apr 7, 2026Apr 25, 2026May 13, 2026Jun 1, 2026

Evaluation results

MethodLinks
DFLAREConcurrency=322026.061,796.9———
DFLAREConcurrency=162026.061,672.1———
DFLAREConcurrency=82026.061,650.2———
DFLAREConcurrency=42026.061,538.6———
BaselineConcurrency=162026.061,273.7———
BaselineConcurrency=322026.061,212———
DFlashConcurrency=82026.061,178.7———
DFlashConcurrency=162026.061,174.9———
BaselineConcurrency=82026.061,158.2———
DFlashConcurrency=322026.061,149.9———
DFlashConcurrency=42026.061,142.9———
BaselineConcurrency=42026.06620.9———
DFlashConcurrency=12026.06561.1———
DFLAREConcurrency=12026.06554.5———
BaselineConcurrency=12026.06163.6———
MoE-SpAc2026.0228.2416.93——
llama.cppspeculative decoding=true2026.0218.7824.01——
llama.cppspeculative decoding=false2026.0214.7430.57——
HybriMoE2026.0211.9638.18——
DDTree-oracleBackbone=Qwen3-4B, Temperature=0.02026.069.32—4.97256
CADDTREEBackbone=Qwen3-4B, Temperature=0.02026.069.32—5.01224
CADDTREEBackbone=Qwen3-8B, Temperature=0.02026.069.2—5.11172
DDTree-oracleBackbone=Qwen3-8B, Temperature=0.02026.068.98—5.14128
Fate2026.027.8260.24——
DFlashBackbone=Qwen3-4B, Temperature=0.02026.066.64—6.82—
DFlashBackbone=Qwen3-8B, Temperature=0.02026.066.48—7—
SP-MoE2026.024.9886.93——
MoE-Infinity2026.023.63119.53——
vLLM2026.022.63158.46——
Mixtral Offload2026.022.48177.87——
Accelerate2026.021.15393.6——
ARBackbone=Qwen3-4B, Temperature=0.02026.06——30.87—
ARBackbone=Qwen3-8B, Temperature=0.02026.06——31.1—