fixed a bug with drafting tokens

LostRuins · LostRuins · commit 10d4fc637d05 · 2024-12-23T11:36:08.000+08:00
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)
                 {

Original file line number	Diff line number	Diff line change
`@@ -3611,7 +3611,7 @@ generation_outputs gpttype_generate(const generation_inputs inputs)`
`3611`	`3611`	`{`
`3612`	`3612`	`logits_to_sample = draft_results.drafted_amount;`
`3613`	`3613`	`}`
`3614`		`- while(logits_sampled<logits_to_sample && remaining_tokens>0 && !abort_draft)`
	`3614`	`+ while(logits_sampled<logits_to_sample && remaining_tokens>0 && !abort_draft && !early_abort)`
`3615`	`3615`	`{`
`3616`	`3616`	`if(logits_sampled>0)`
`3617`	`3617`	`{`