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

streamline output console log (+1 squashed commits)

Browse source 
Squashed commits:

[ca474bdd] streamline output console log
This commit is contained in:
Concedo 2025-03-13 14:58:08 +08:00
parent 16137f4281
commit 1ef41c2124
2 changed files with 13 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

5
examples/llava/clip.cpp
View file

@ -3171,6 +3171,11 @@ bool clip_model_quantize(const char * fname_inp, const char * fname_out, const i
{ {
quantize = false; quantize = false;
} }
// //temp fix for gemma3
// if(name.find("ffn_up.weight") != std::string::npos)
// {
// quantize = false;
// }
} }
if (quantize) { if (quantize) {

9
gpttype_adapter.cpp
View file

@ -3993,7 +3993,14 @@ generation_outputs gpttype_generate(const generation_inputs inputs)
float pt2 = (time2*1000.0/(realnpredict<=0?1:realnpredict)); float pt2 = (time2*1000.0/(realnpredict<=0?1:realnpredict));
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); if(debugmode==1)
{
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);
}
else
{
printf("\n[%s] CtxLimit:%d/%d, Amt:%d/%d, Init:%.2fs, Process:%.2fs (%.2fT/s), Generate:%.2fs (%.2fT/s), Total:%.2fs",get_timestamp_str().c_str(),(int)current_context_tokens.size(),(int)nctx, realnpredict, kcpp_data->n_predict, time0, time1, ts1, time2, ts2, (time1 + time2));
}
if(debugmode==1 && !is_quiet && (draft_successes+draft_failures)>0) if(debugmode==1 && !is_quiet && (draft_successes+draft_failures)>0)
{ {
printf("\n(Draft Results - Success:%d, Failure:%d)",draft_successes,draft_failures); printf("\n(Draft Results - Success:%d, Failure:%d)",draft_successes,draft_failures);