List your company

Mathematical Reasoning on GSM8K (Inference Latency & Throughput)

1,163Latency (NoKV) (ms)NoKV
1,144.161,271.331,398.51,525.67Jun 7, 2026

Evaluation results

MethodLinks
NoKVModel=Qwen2.5-7B Instruct2026.061,163—————
NoKVModel=Falcon3-7B Instruct2026.061,170—————
NoKVModel=Mistral-7B Instruct2026.061,262—————
NoKVModel=Llama-3-8B Instruct2026.061,283—————
NoKVModel=Falcon3-10B Instruct2026.061,634—————
RKSCModel=Qwen2.5-7B Instruct2026.06——515—2.2632
RKSCModel=Mistral-7B Instruct2026.06——581—2.1730
RKSCModel=Falcon3-7B Instruct2026.06——376—3.1150
RKSCModel=Falcon3-10B Instruct2026.06——473—3.4657
RKSCModel=Llama-3-8B Instruct2026.06——464—2.7665
vLLMModel=Qwen2.5-7B Instruct2026.06—652—1.78——
vLLMModel=Mistral-7B Instruct2026.06—689—1.83——
vLLMModel=Falcon3-7B Instruct2026.06—650—1.8——
vLLMModel=Falcon3-10B Instruct2026.06—898—1.82——
vLLMModel=Llama-3-8B Instruct2026.06—710—1.81——