fixed a bug with drafting tokens

LostRuins · Dec 23, 2024 · 10d4fc6 · 10d4fc6
1 parent fd5100c
commit 10d4fc6
Showing 1 changed file with 1 addition and 1 deletion.
diff --git a/gpttype_adapter.cpp b/gpttype_adapter.cpp
@@ -3611,7 +3611,7 @@ generation_outputs gpttype_generate(const generation_inputs inputs)
             {
                 logits_to_sample = draft_results.drafted_amount;
             }
-            while(logits_sampled<logits_to_sample && remaining_tokens>0 && !abort_draft)
+            while(logits_sampled<logits_to_sample && remaining_tokens>0 && !abort_draft && !early_abort)
             {
                 if(logits_sampled>0)
                 {