List your company

Inference Efficiency on Alpaca

102.44Latency (ms/tok)Baseline
101.7236106.5593111.395116.2307Aug 14, 2025

Evaluation results

MethodLinks
BaselineBackbone=Llama3-8B-Instruct, Sequence length=512, Batch size=12025.08102.449.7616.1
CAABackbone=Llama3-8B-Instruct, Sequence length=512, Batch size=12025.08110.599.0416.12
ReFTBackbone=Llama3-8B-Instruct, Sequence length=512, Batch size=12025.08115.728.6416.26
MSRSBackbone=Llama3-8B-Instruct, Sequence length=512, Batch size=12025.08120.358.3816.29