List your company

Single-stream decode on SmolLM2-135M

8.65Latency (ms/token)vLLM cudagraph
7.713614.034320.35526.6757Jun 8, 2026

Evaluation results

MethodLinks
vLLM cudagraphGPU=H100, dtype=float322026.068.651.65
vLLM enforce-eagerGPU=H100, dtype=float322026.0613.511.06
AMKGPU=H100, Comparison Baseline=vLLM cudagraph, Optimization=earlier AMK GEMV2026.0614.3—
AMKGPU=H100, Comparison Baseline=vLLM enforce-eager, Optimization=earlier AMK GEMV2026.0614.3—
AMKGPU=A100, Comparison Baseline=vLLM enforce-eager, Optimization=earlier AMK GEMV2026.0615.592.06
vLLM enforce-eagerGPU=A100, dtype=float322026.0632.06—