send partial stop string when <EOG> is reached

matteoserva · matteoserva · commit a39eac1117ef · 2025-08-22T11:10:03.000+02:00
diff --git a/tools/server/server.cpp b/tools/server/server.cpp
@@ -2360,7 +2360,7 @@ 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
`@@ -2360,7 +2360,7 @@ struct server_context {`
`2360`	`2360`	`slot.generated_text.begin() + pos + stop_pos,`
`2361`	`2361`	`slot.generated_text.end());`
`2362`	`2362`	`pos = std::min(slot.n_sent_text, slot.generated_text.size());`
`2363`		`- } else if (slot.has_next_token) {`
	`2363`	`+ } else if (slot.has_next_token && !llama_vocab_is_eog(vocab, result.tok) ) {`
`2364`	`2364`	`stop_pos = slot.find_stopping_strings(str_test, token_str.size(), false);`
`2365`	`2365`	`send_text = stop_pos == std::string::npos;`
`2366`	`2366`	`}`