add quick hack to render only some special tokens

Author: ngxson

src/llama-vocab.cpp

@@ -2330,6 +2330,13 @@ void llama_vocab::impl::load(llama_model_loader & ml, const LLM_KV & kv) {
             }
         }
 
+        // @ngxson : quick hack for gpt-oss
+        for (const auto & t : token_to_id) {
+            if (t.first == "<|channel|>" || t.first == "<|message|>") {
+                id_to_token[t.second].attr = LLAMA_TOKEN_ATTR_NORMAL;
+            }
+        }
+
         // sanity checks
         if (special_eos_id != LLAMA_TOKEN_NULL && special_eog_ids.count(special_eos_id) == 0) {
             special_eog_ids.insert(special_eos_id);

tools/server/public/index.html.gz

@@ -61,20 +61,22 @@ export default function ChatMessage({
     if (msg.content === null || msg.role !== 'assistant') {
       return { content: msg.content };
     }
+    const REGEX_THINK_OPEN = /<think>|<\|channel\|>analysis<\|message\|>/;
+    const REGEX_THINK_CLOSE = /<\/think>|<\|channel\|>final<\|message\|>/;
     let actualContent = '';
     let thought = '';
     let isThinking = false;
-    let thinkSplit = msg.content.split('<think>', 2);
+    let thinkSplit = msg.content.split(REGEX_THINK_OPEN, 2);
     actualContent += thinkSplit[0];
     while (thinkSplit[1] !== undefined) {
       // <think> tag found
-      thinkSplit = thinkSplit[1].split('</think>', 2);
+      thinkSplit = thinkSplit[1].split(REGEX_THINK_CLOSE, 2);
       thought += thinkSplit[0];
       isThinking = true;
       if (thinkSplit[1] !== undefined) {
         // </think> closing tag found
         isThinking = false;
-        thinkSplit = thinkSplit[1].split('<think>', 2);
+        thinkSplit = thinkSplit[1].split(REGEX_THINK_OPEN, 2);
         actualContent += thinkSplit[0];
       }
     }