url/colibri
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

run_text: TOKENS=1 env to dump generated token ids for path A/B comparison

Browse Source 
Opt-in (default off): prints the generated token ids to stderr after
generation, for exact comparison across decode paths (e.g. the S=1
resident pipeline vs the CPU path). Used to verify the pipe_layer_sparse
gate relaxation is token-exact vs the CPU decode path (see #273).
This commit is contained in:
woolcoxm
2026-07-15 16:31:18 -04:00
parent a55cdfd4b8
commit e56d483a15
1 changed files with 8 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
c/glm.c
+8
View File
@@ -3916,6 +3916,14 @@ static void run_text(Model *m, const char *snap, const char *prompt, int ngen){
for(int i=0;i<4;i++) printf(" %-42s %5.1f%% (%lld/%lld)\n", nm[i],
la_tot[i]?100.0*la_hit[i]/la_tot[i]:0.0, (long long)la_hit[i], (long long)la_tot[i]);
}
/* TOKENS=1: dump the generated token ids (newline-separated) to stderr,
* for exact A/B comparison across decode paths (e.g. resident vs CPU).
* The ids are all[np .. np+produced-1]. */
if(getenv("TOKENS") && atoi(getenv("TOKENS"))){
fprintf(stderr,"[TOKENS] %d generated:",produced);
for(int i=np;i<np+produced;i++) fprintf(stderr," %d",all[i]);
fprintf(stderr,"\n");
}
free(pids); free(all);
usage_save(m);
}