[Bugfix] Improve GLM4 MoE Reasoning Parser's is_reasoning_end Condition

[frankwang28]

Purpose

When testing GLM-4.5 in streaming mode with both reasoning and tool calls enabled, it was noticed that after the first assistant response, reasoning was not being parsed and added to the reasoning_content properly.

This is due to calling is_reasoning_end on all the pervious prompt tokens .

GLM-4.5's chat template removes the reasoning content, but still appends <think><\think> after the assistant token for previous interactions (interactive example). Thus, when checking for just the <\think> token on all prompt tokens, the check would return true as the previous turn's <\think> token is present.

Thus, we must add the case that if there is an <|assistant|> token present, check that there is a <\think> token after the last <|assistant|> token. This can equivalently be accomplished by looping backwards over the prompt tokens and returning false upon encountering the <|assistant|> token or true upon encountering a <\think> token.

Test Plan

Added automated tests (tests/reasoning/test_glm4_moe_reasoning_parser.py).
Additionally performed manual test with the following curl command:

curl --location 'http://0.0.0.0:8000/v1/chat/completions' \
--header 'Content-Type: application/json' \
--data '{
    "temperature": 0.7,
    "max_tokens": 5,
    "model": "zai-org/GLM-4.5-Air-FP8",
    "stream": true,
    "tools": [
        {
            "type": "function",
            "function": {
                "name": "get_weather",
                "description": "This function gets the weather of a certain location.",
                "parameters": {
                    "type": "object",
                    "properties": {
                        "location": {
                            "type": "string",
                            "description": "weather location"
                        }
                    },
                    "required": [
                        "location"
                    ]
                }
            }
        }
    ],
    "messages": [
        {
            "role": "user",
            "content": "What'\''s the current weather in Vancouver?"
        },
        {
            "role": "assistant",
            "content": "\n\n",
            "tool_calls": [
                {
                    "id": "12345",
                    "type": "function",
                    "function": {
                        "name": "get_weather",
                        "arguments": "{\"location\": \"Vancouver\"}"
                    }
                }
            ],
            "reasoning_content": "The user is asking for the current weather in Vancouver. I need to search for this information. I'\''ll use the get_weather function to find current weather information for Vancouver.",
            "id": "chatcmpl-12345",
            "cited_sources": null
        },
        {
            "role": "tool",
            "content": "The weather in Vancouver is clear and 17 degrees Celsius."
        }
    ]
}'

Test Result

Automated tests pass.

Curl command results:
Main

data: {"id":"chatcmpl-e09005f6e45949af9e072efe5bacded5","object":"chat.completion.chunk","created":1758448039,"model":"zai-org/GLM-4.5-Air-FP8","choices":[{"index":0,"delta":{"role":"assistant","content":""},"logprobs":null,"finish_reason":null}],"prompt_token_ids":null}

data: {"id":"chatcmpl-e09005f6e45949af9e072efe5bacded5","object":"chat.completion.chunk","created":1758448039,"model":"zai-org/GLM-4.5-Air-FP8","choices":[{"index":0,"delta":{"content":null},"logprobs":null,"finish_reason":null,"token_ids":null}]}

data: {"id":"chatcmpl-e09005f6e45949af9e072efe5bacded5","object":"chat.completion.chunk","created":1758448039,"model":"zai-org/GLM-4.5-Air-FP8","choices":[{"index":0,"delta":{"content":"\n<think>"},"logprobs":null,"finish_reason":null,"token_ids":null}]}

data: {"id":"chatcmpl-e09005f6e45949af9e072efe5bacded5","object":"chat.completion.chunk","created":1758448039,"model":"zai-org/GLM-4.5-Air-FP8","choices":[{"index":0,"delta":{"content":"The"},"logprobs":null,"finish_reason":null,"token_ids":null}]}

data: {"id":"chatcmpl-e09005f6e45949af9e072efe5bacded5","object":"chat.completion.chunk","created":1758448039,"model":"zai-org/GLM-4.5-Air-FP8","choices":[{"index":0,"delta":{"content":" weather"},"logprobs":null,"finish_reason":null,"token_ids":null}]}

data: {"id":"chatcmpl-e09005f6e45949af9e072efe5bacded5","object":"chat.completion.chunk","created":1758448039,"model":"zai-org/GLM-4.5-Air-FP8","choices":[{"index":0,"delta":{"content":" function"},"logprobs":null,"finish_reason":"length","stop_reason":null,"token_ids":null}]}

data: [DONE]

This PR

data: {"id":"chatcmpl-c7b139841cd146b19cebdc073814c4fa","object":"chat.completion.chunk","created":1758478817,"model":"zai-org/GLM-4.5-Air-FP8","choices":[{"index":0,"delta":{"role":"assistant","content":""},"logprobs":null,"finish_reason":null}],"prompt_token_ids":null}

data: {"id":"chatcmpl-c7b139841cd146b19cebdc073814c4fa","object":"chat.completion.chunk","created":1758478817,"model":"zai-org/GLM-4.5-Air-FP8","choices":[{"index":0,"delta":{"content":"\n"},"logprobs":null,"finish_reason":null,"token_ids":null}]}

data: {"id":"chatcmpl-c7b139841cd146b19cebdc073814c4fa","object":"chat.completion.chunk","created":1758478817,"model":"zai-org/GLM-4.5-Air-FP8","choices":[{"index":0,"delta":{"reasoning_content":"The"},"logprobs":null,"finish_reason":null,"token_ids":null}]}

data: {"id":"chatcmpl-c7b139841cd146b19cebdc073814c4fa","object":"chat.completion.chunk","created":1758478817,"model":"zai-org/GLM-4.5-Air-FP8","choices":[{"index":0,"delta":{"reasoning_content":" function"},"logprobs":null,"finish_reason":null,"token_ids":null}]}

data: {"id":"chatcmpl-c7b139841cd146b19cebdc073814c4fa","object":"chat.completion.chunk","created":1758478817,"model":"zai-org/GLM-4.5-Air-FP8","choices":[{"index":0,"delta":{"reasoning_content":" returned"},"logprobs":null,"finish_reason":"length","stop_reason":null,"token_ids":null}]}

data: [DONE]

---

Essential Elements of an Effective PR Description Checklist

• The purpose of the PR, such as "Fix some issue (link existing issues this PR will resolve)".
• The test plan, such as providing test command.
• The test results, such as pasting the results comparison before and after, or e2e results
• (Optional) The necessary documentation update, such as updating supported_models.md and examples for a new model.
• (Optional) Release notes update. If your change is user facing, please update the release notes draft in the Google Doc.

[gemini-code-assist]

Code Review

This pull request effectively resolves a bug in the GLM-4 MoE reasoning parser for multi-turn streaming scenarios. The logic change in is_reasoning_end is correct and well-supported by the comprehensive new test suite. I have one suggestion to enhance the robustness of the parser by ensuring all necessary special tokens are validated during initialization.

[chaunceyjiang]

/cc @zRzRzRzRzRzRzR PTAL.

[chaunceyjiang]

Thanks~