fixes

Author: aldehir

common/parsers/harmony.cpp

@@ -9,6 +9,7 @@ void harmony_msg_parser::parse() {
     //builder.try_parse_reasoning("<|channel|>analysis<|message|>", "<|start|>assistant<|channel|>final<|message|>");
     //builder.add_content(builder.consume_rest());
     channel();
+    builder.add_content(builder.consume_rest());
 }
 
 void harmony_msg_parser::channel() {

src/llama-vocab.cpp

@@ -2342,7 +2342,7 @@ void llama_vocab::impl::load(llama_model_loader & ml, const LLM_KV & kv) {
 
         // @ngxson : quick hack for gpt-oss, always render these tokens
         for (const auto & t : token_to_id) {
-            if (t.first == "<|channel|>" || t.first == "<|message|>" || t.first == "<|start|>") {
+            if (t.first == "<|channel|>" || t.first == "<|message|>" || t.first == "<|start|>" || t.first == "<|constrain|>") {
                 id_to_token[t.second].attr = LLAMA_TOKEN_ATTR_USER_DEFINED;
             }
         }
@@ -2389,6 +2389,7 @@ void llama_vocab::impl::load(llama_model_loader & ml, const LLM_KV & kv) {
 
             if (has_return && has_call && has_end) {
                 special_eog_ids.erase(end_id);
+                id_to_token[end_id].attr = LLAMA_TOKEN_ATTR_USER_DEFINED;
                 LLAMA_LOG_WARN("%s: special_eog_ids contains both '<|return|>' and '<|call|>' tokens, removing '<|end|>' token from EOG list\n", __func__);
             }
         }

tests/test-chat.cpp

@@ -1520,7 +1520,7 @@ static void test_template_output_parsers() {
         assert_msg_equals(simple_assist_msg("", "I'm\nthinking"),
             common_chat_parse(
                 "<|channel|>analysis<|message|>I'm\nthinking<|end|>"
-                "<|start|>assistant<|channel|><|final|>",
+                "<|start|>assistant<|channel|>final",
                 /* is_partial= */ true,
                 {
                     /* .format = */ COMMON_CHAT_FORMAT_GPT_OSS,
@@ -1529,7 +1529,7 @@ static void test_template_output_parsers() {
         assert_msg_equals(simple_assist_msg("", "I'm\nthinking"),
             common_chat_parse(
                 "<|channel|>analysis<|message|>I'm\nthinking<|end|>"
-                "<|start|>assistant<|channel|><|final|><|message|>",
+                "<|start|>assistant<|channel|>final<|message|>",
                 /* is_partial= */ true,
                 {
                     /* .format = */ COMMON_CHAT_FORMAT_GPT_OSS,
@@ -1601,7 +1601,7 @@ static void test_template_output_parsers() {
         assert_msg_equals(simple_assist_msg("", "I'm\nthinking"),
             common_chat_parse(
                 "<|channel|>analysis<|message|>I'm\nthinking<|end|>"
-                "<|start|>assistant<|channel|>commentary ",
+                "<|start|>assistant<|channel|>commentary",
                 /* is_partial= */ true,
                 {
                     /* .format = */ COMMON_CHAT_FORMAT_GPT_OSS,