From 6c855db82c36b4d86b551ae95624d8eaf9f16ba5 Mon Sep 17 00:00:00 2001
From: Xinyuan Tong <115166877+JustinTong0323@users.noreply.github.com>
Date: Fri, 22 Aug 2025 08:24:25 +0800
Subject: [PATCH] Revert "bugfix: Fix output_ids extraction in
 detokenizer_manager" (#9467)

---
 python/sglang/srt/entrypoints/context.py      | 19 ++++++++++++++++++-
 .../srt/managers/detokenizer_manager.py       |  2 +-
 2 files changed, 19 insertions(+), 2 deletions(-)

diff --git a/python/sglang/srt/entrypoints/context.py b/python/sglang/srt/entrypoints/context.py
index 9b0791101..66f58200f 100644
--- a/python/sglang/srt/entrypoints/context.py
+++ b/python/sglang/srt/entrypoints/context.py
@@ -1,5 +1,5 @@
 # SPDX-License-Identifier: Apache-2.0
-# Copied from vLLM: https://github.com/zyongye/vllm/blob/6a70830065701b163e36a86fd331b41b5feac401/vllm/entrypoints/context.py
+# Copied from vLLM
 import json
 import logging
 from abc import ABC, abstractmethod
@@ -83,6 +83,14 @@ class HarmonyContext(ConversationContext):
         if isinstance(output, dict) and "output_ids" in output:
             output_token_ids = output["output_ids"]
 
+            # TODO: REMOVE here:
+            # Very hacky, find the first occurrence of token 200006 and cut from there
+            try:
+                start_index = output_token_ids.index(200006)
+                output_token_ids = output_token_ids[start_index:]
+            except ValueError:
+                pass
+
             for token_id in output_token_ids:
                 self.parser.process(token_id)
             output_msgs = self.parser.messages
@@ -190,6 +198,15 @@ class StreamingHarmonyContext(HarmonyContext):
             # RequestOutput from SGLang with outputs
             output_token_ids = output["output_ids"]
 
+            # TODO: REMOVE here:
+            # Very hacky, find the first occurrence of token 200006 and cut from there
+            # Find the first occurrence of token 200006 and cut from there
+            try:
+                start_index = output_token_ids.index(200006)
+                output_token_ids = output_token_ids[start_index:]
+            except ValueError:
+                pass
+
             for token_id in output_token_ids:
                 self.parser.process(token_id)
 
diff --git a/python/sglang/srt/managers/detokenizer_manager.py b/python/sglang/srt/managers/detokenizer_manager.py
index 34a29ec17..29757b4b2 100644
--- a/python/sglang/srt/managers/detokenizer_manager.py
+++ b/python/sglang/srt/managers/detokenizer_manager.py
@@ -216,7 +216,7 @@ class DetokenizerManager:
             rids=recv_obj.rids,
             finished_reasons=recv_obj.finished_reasons,
             output_strs=output_strs,
-            output_ids=recv_obj.output_ids,
+            output_ids=recv_obj.decode_ids,
             prompt_tokens=recv_obj.prompt_tokens,
             completion_tokens=recv_obj.completion_tokens,
             cached_tokens=recv_obj.cached_tokens,