diff --git a/gpttype_adapter.cpp b/gpttype_adapter.cpp index 7bba001cc..8db770377 100644 --- a/gpttype_adapter.cpp +++ b/gpttype_adapter.cpp @@ -2088,7 +2088,7 @@ generation_outputs gpttype_generate(const generation_inputs inputs, generation_o float pt2 = (time2*1000.0/(realnpredict==0?1:realnpredict)); float ts2 = (1000.0/pt2); float tokens_per_second = (realnpredict == 0 ? 0 : realnpredict / (time1 + time2)); - printf("\nContextLimit: %d/%d, Processing:%.2fs (%.1fms/T = %.2fT/s), Generation:%.2fs (%.1fms/T = %.2fT/s), Total:%.2fs (%.1fms/T = %.2fT/s)",current_context_tokens.size(),nctx, time1, pt1, ts1, time2, pt2, ts2, (time1 + time2), (1000.0f/tokens_per_second) , tokens_per_second); + printf("\nCtxLimit: %d/%d, Process:%.2fs (%.1fms/T = %.2fT/s), Generate:%.2fs (%.1fms/T = %.2fT/s), Total:%.2fs (%.2fT/s)",current_context_tokens.size(),nctx, time1, pt1, ts1, time2, pt2, ts2, (time1 + time2), tokens_per_second); fflush(stdout); output.status = 1; generation_finished = true;