vrr/koboldcpp
2
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2025-09-10 09:04:36 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

added draft results

Browse source
This commit is contained in:
Concedo 2025-03-10 22:03:20 +08:00
parent 3a406b37a7
commit b0541f3652
1 changed files with 9 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

9
gpttype_adapter.cpp
View file

@ -3357,6 +3357,8 @@ generation_outputs gpttype_generate(const generation_inputs inputs)
speculative_draft_result draft_results; //only use if drafting was used speculative_draft_result draft_results; //only use if drafting was used
bool draft_used = false; bool draft_used = false;
int draft_successes = 0;
int draft_failures = 0;
time0 = timer_check(); time0 = timer_check();
timer_start(); timer_start();
@ -3665,7 +3667,10 @@ generation_outputs gpttype_generate(const generation_inputs inputs)
} }
if(draftedid!=id) //draft mismatch, abort if(draftedid!=id) //draft mismatch, abort
{ {
draft_failures += 1;
abort_draft = true; abort_draft = true;
} else {
draft_successes += 1;
} }
} }
@ -3983,6 +3988,10 @@ generation_outputs gpttype_generate(const generation_inputs inputs)
float ts2 = (1000.0/pt2); float ts2 = (1000.0/pt2);
float tokens_per_second = (realnpredict <= 0 ? 0 : realnpredict / (time1 + time2)); float tokens_per_second = (realnpredict <= 0 ? 0 : realnpredict / (time1 + time2));
printf("\n[%s] CtxLimit:%d/%d, Amt:%d/%d, Init:%.2fs, Process:%.2fs (%.1fms/T = %.2fT/s), Generate:%.2fs (%.1fms/T = %.2fT/s), Total:%.2fs (%.2fT/s)",get_timestamp_str().c_str(),(int)current_context_tokens.size(),(int)nctx, realnpredict, kcpp_data->n_predict, time0, time1, pt1, ts1, time2, pt2, ts2, (time1 + time2), tokens_per_second); printf("\n[%s] CtxLimit:%d/%d, Amt:%d/%d, Init:%.2fs, Process:%.2fs (%.1fms/T = %.2fT/s), Generate:%.2fs (%.1fms/T = %.2fT/s), Total:%.2fs (%.2fT/s)",get_timestamp_str().c_str(),(int)current_context_tokens.size(),(int)nctx, realnpredict, kcpp_data->n_predict, time0, time1, pt1, ts1, time2, pt2, ts2, (time1 + time2), tokens_per_second);
if(debugmode==1 && !is_quiet && (draft_successes+draft_failures)>0)
{
printf("\n(Draft Results - Success:%d, Failure:%d)",draft_successes,draft_failures);
}
fflush(stdout); fflush(stdout);
output.status = 1; output.status = 1;
int finaltokcount = (int)current_context_tokens.size()-realnpredict; int finaltokcount = (int)current_context_tokens.size()-realnpredict;