@@ -1230,7 +1230,7 @@ struct common_speculative_session::impl {
12301230 return !draft.empty ();
12311231 }
12321232
1233- void clear_draft () {
1233+ void leave_draft_state () {
12341234 draft.clear ();
12351235 spec_ckpt_n_denials = 0 ;
12361236 }
@@ -1241,19 +1241,19 @@ struct common_speculative_session::impl {
12411241 const int n_draft_max) {
12421242 if (spec == nullptr ) {
12431243 // no implementation, nothing to do
1244- clear_draft ();
1244+ leave_draft_state ();
12451245 return draft;
12461246 }
12471247
12481248 if (n_draft_max == 0 ) {
1249- clear_draft ();
1249+ leave_draft_state ();
12501250 return draft;
12511251 }
12521252 if (params_spec.use_checkpoints && spec_ckpt_n_denials > 1 ) {
12531253 // We shouldn't get two denials.
12541254 LOG_WRN (" %s: #tokens=%zu, spec_ckpt_n_denials=%d, id_last=%d, #draft=%zu\n " , __func__,
12551255 cached_text_tokens.size (), spec_ckpt_n_denials, id_last, draft.size ());
1256- clear_draft ();
1256+ leave_draft_state ();
12571257 return draft;
12581258 }
12591259
@@ -1262,6 +1262,7 @@ struct common_speculative_session::impl {
12621262 if (draft.empty ()) {
12631263 // switch to non-draft inference
12641264 LOG_DBG (" %s: draft of length 0 after denied checkpoint\n " , __func__);
1265+ leave_draft_state ();
12651266 return draft;
12661267 }
12671268 // we use the shortened draft of previous speculation
@@ -1274,7 +1275,7 @@ struct common_speculative_session::impl {
12741275 draft = common_speculative_draft (spec, params_spec, cached_text_tokens, id_last);
12751276 LOG_DBG (" draft: id_last=%d, #draft=%zu\n " , id_last, draft.size ());
12761277 if (draft.empty ()) {
1277- clear_draft ();
1278+ leave_draft_state ();
12781279 return draft;
12791280 }
12801281 }
@@ -1299,15 +1300,15 @@ struct common_speculative_session::impl {
12991300
13001301 if (params_spec.n_min > (int ) draft.size ()) {
13011302 LOG_DBG (" ignoring small draft: %d < %d\n " , (int ) draft.size (), params_spec.n_min );
1302- clear_draft ();
1303+ leave_draft_state ();
13031304 return draft;
13041305 }
13051306
13061307 if (do_checkpoint) {
13071308 const size_t n = callback.create_checkpoint ();
13081309 if (n == 0 ) {
13091310 LOG_WRN (" %s: checkpoint creation failed (#tokens=%zu)\n " , __func__, cached_text_tokens.size ());
1310- clear_draft ();
1311+ leave_draft_state ();
13111312 return draft;
13121313 }
13131314 spec_ckpt_size_part = n;
@@ -1398,7 +1399,7 @@ struct common_speculative_session::impl {
13981399 return ;
13991400 }
14001401
1401- clear_draft ();
1402+ leave_draft_state ();
14021403
14031404 spec_has_ckpt = false ;
14041405 spec_ckpt_size_part = 0 ;
0 commit comments