Fix error in deepseek parser

[tdene]

• I, the PR author, have personally reviewed every line of this PR.

What does this PR do ?

Comparison of vllm and Megatron-LM reasoning parser.

MLM path: https://github.com/NVIDIA/Megatron-LM/blob/main/megatron/core/tokenizers/text/parsers/deepseek_r1_reasoning_parser.py
vllm paths: https://github.com/vllm-project/vllm/blob/main/vllm/reasoning/deepseek_r1_reasoning_parser.py https://github.com/vllm-project/vllm/blob/main/vllm/reasoning/basic_parsers.py

MLM code:

        if "</think>" in text:
            if "<think>" in text:
                # Strip the <think> prefix (it might not be present if it was part of the prompt)
                pre_text, text = text.split("<think>", maxsplit=1)
            else:
                pre_text = ""
            reasoning_content, remaining_text = text.split("</think>", maxsplit=1)
            return pre_text + remaining_text, {'reasoning': reasoning_content}
        else:
            return text, {}

vllm code:

        model_output_parts = model_output.partition(self.start_token)
        model_output = (
            model_output_parts[2] if model_output_parts[1] else model_output_parts[0]
        )

        # For models that may not generate start token,
        # assume the reasoning content is always at the start.
        if self.end_token not in model_output:
            return model_output, None
        else:
            reasoning, _, content = model_output.partition(self.end_token)
            # If generation stops right after end-of-think, return null content
            final_content = content or None
            return reasoning, final_content

NOTE: In this table, the tuple represents (reasoning, content).

+---------------------------+--------------------+--------------------+-------+
| input                     |          vLLM      | Megatron-LM        |
+---------------------------+--------------------+--------------------+-------+
| 'answer'                  | ('answer', None)   | (None, 'answer')   |
| '<think>r'                | ('r', None)        | (None, '<think>r') |
| 'pre<think>r</think>post' | ('r', 'post')      | ('r', 'prepost')   |
| 'r</think>a<think>b'      | ('b', None)        | CRASH (ValueError) |
+---------------------------+--------------------+--------------------+-------+

In a nutshell:

• vllm assumes that all rollouts that use a reasoning parser are CoT, and there is never a case where we will skip reasoning and go straight to non-reasoning content.
• vllm throws away everything before the first explicit <think> tag.
• vllm does not need an explicit </think> tag,

⚠️ For major changes (either in lines of code or in its impact), please make sure to first share a design doc with the team. If you're unsure what's the best way to do so, contact @NVIDIA/mcore-oncall.

Issue tracking

For PRs from open-source community contributors:

• New features: a linked issue is required. Please open a feature request and reference it here before submitting the PR.
• Small updates (bug fixes, minor improvements): a linked issue is recommended and will accelerate the PR review process.

Linked issue:

Contribution process

Pre-checks

• I have added relevant unit tests
• I have added relevant functional tests
• I have added proper typing to my code Typing guidelines
• I have added relevant documentation
• I have run the autoformatter.sh on my PR

Code review

Feel free to message or comment @NVIDIA/mcore-oncall to help accelerate your merge into main. The less complex your PR is, the faster it will be approved and merged!

All PRs start as draft. If you open a non-draft PR, it will be automatically converted to draft.

Step 1: Mark PR as "Ready for Review"

1. When your PR is ready, click Ready for Review.
2. An oncall reviewer is auto-assigned and expert reviewers are notified based on your changes.
   • Some PRs may jump straight to step 2. This is determined by .github/CODEOWNERS.

⚠️ Only mark as ready once merge-conflicts are resolved and the CI is passing.
Final Review might get declined if these requirements are not fulfilled.

Step 2: Final Review

For PRs that change megatron/core, once all expert reviewers have approved, the Final Review label is applied automatically and final reviewers are assigned.

For PRs outside megatron/core, this step is skipped.

Step 3: Approved

Once all required reviewers have approved, the Approved label is applied automatically.

Merge

Any member of mcore-engineers will be able to merge your PR.

[copy-pr-bot]

Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually.

Contributors can view more details about this message here.

[ericharper]

/ok to test 4a0fb9c

[ericharper]

/ok to test 4a0fb9c

[svcnvidia-nemo-ci]

🔄 Merge queue validation started!

You can track the progress here: https://github.com/NVIDIA/Megatron-LM/actions/runs/27027461512

[svcnvidia-nemo-ci]

🔄 Merge queue validation started!

You can track the progress here: https://github.com/NVIDIA/Megatron-LM/actions/runs/27037180102