issue/523 - switched to 1.22 interface

[wooway777]

resolves #523

硬件型号：mlu 590 +phytum cpu

单卡

python examples/jiuge.py --cambricon --model_path=/data/pepe/9G7B_MHA/ --max_new_tokens=1024
 Generation completed in 2108.3 ms
 Batchsize=1  Per_Batch_Input_Len=13  Per_Batch_New_Tokens=40

 Prefill TTFT: 0.11ms  Throughput: 118.66tok/s

 Decode  Avg ITL: 51.25ms   Throughput: 19.51tok/s

分布式由于通信起不来跑不了，应该不是代码问题

基准测试

python test/bench/test_benchmark.py --cambricon /data/pepe/9G7B_MHA/ --bench ceval --subject middle_school_mathematics --num_samples 100 --backend cpp --ndev 1

Overall 成绩: 61/100 = 61.00%
Total Latency: 798.5647848986555 seconds
Total Tokens Processed: 25442 tokens
Overall Throughput: 31.86 tokens/s

[wooway777]

单卡

基准

分布式

通信