Energy Savings:71.4x· Arithmetic operations on 7nmLatency Speedup:4.1x· 70B model sizeMemory Reduction:3.55x· 3B model sizePerplexity (PPL):9.91· WikiText2 / C4
Energy Savings:71.4x· Arithmetic operations on 7nmLatency Speedup:4.1x· 70B model sizeMemory Reduction:3.55x· 3B model sizePerplexity (PPL):9.91· WikiText2 / C4