send partial stop string when <EOG> is reached

matteoserva · matteoserva · commit d3dd9e0e9b9a · 2025-09-07T18:17:16.000+02:00
diff --git a/tools/server/server.cpp b/tools/server/server.cpp
@@ -2541,7 +2541,7 @@ struct server_context {
                     slot.generated_text.begin() + pos + stop_pos,
                     slot.generated_text.end());
                 pos = std::min(slot.n_sent_text, slot.generated_text.size());
-            } else if (slot.has_next_token) {
+            } else if (slot.has_next_token && !llama_vocab_is_eog(vocab, result.tok) ) {
                 stop_pos = slot.find_stopping_strings(str_test, token_str.size(), false);
                 send_text = stop_pos == std::string::npos;
             }

Original file line number	Diff line number	Diff line change
`@@ -2541,7 +2541,7 @@ struct server_context {`
`2541`	`2541`	`slot.generated_text.begin() + pos + stop_pos,`
`2542`	`2542`	`slot.generated_text.end());`
`2543`	`2543`	`pos = std::min(slot.n_sent_text, slot.generated_text.size());`
`2544`		`- } else if (slot.has_next_token) {`
	`2544`	`+ } else if (slot.has_next_token && !llama_vocab_is_eog(vocab, result.tok) ) {`
`2545`	`2545`	`stop_pos = slot.find_stopping_strings(str_test, token_str.size(), false);`
`2546`	`2546`	`send_text = stop_pos == std::string::npos;`
`2547`	`2547`	`}`