Struct MiniMaxAppendThinkParser 

pub struct MiniMaxAppendThinkParser { /* private fields */ }

MiniMax Append-Think Reasoning Parser.

The MiniMax model starts generating reasoning content immediately WITHOUT emitting a <think> opener in its output. SGLang’s MiniMaxAppendThinkDetector and vLLM’s MiniMaxM2AppendThinkReasoningParser both handle this by simply prepending <think> to the emitted text and classifying the whole stream as normal_text/content — neither extracts reasoning based on a </think> marker. The tag is left inline for downstream consumers that want to render or post-process it.

This parser matches those upstream implementations verbatim: a pass-through with a one-time <think> prefix on the first streamed chunk. Reasoning content is never populated.

References:

• SGLang MiniMaxAppendThinkDetector: https://github.com/sgl-project/sglang/blob/main/python/sglang/srt/parser/reasoning_parser.py
• vLLM MiniMaxM2AppendThinkReasoningParser: https://github.com/vllm-project/vllm/blob/main/vllm/reasoning/minimax_m2_reasoning_parser.py

Implementations

impl MiniMaxAppendThinkParser

pub fn new() -> Self

Trait Implementations

impl Debug for MiniMaxAppendThinkParser

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for MiniMaxAppendThinkParser

fn default() -> MiniMaxAppendThinkParser

Returns the “default value” for a type.

impl ReasoningParser for MiniMaxAppendThinkParser

fn detect_and_parse_reasoning( &mut self, text: &str, _token_ids: &[u32], ) -> ParserResult

Parses a standalone, non-streaming input chunk. Implementations may reset or ignore internal streaming state and should return the split of normal vs reasoning text for this complete input. Marker tokens must not be included in either output.

fn parse_reasoning_streaming_incremental( &mut self, text: &str, _token_ids: &[u32], ) -> ParserResult

Parses a streaming chunk and updates internal state. The return value should be the delta: only the newly discovered normal and reasoning text attributable to this chunk (not the cumulative totals). Marker tokens must not be included in either output.

fn finish_reasoning_stream(&mut self) -> ParserResult

Finalizes a stream after the last chunk, before parser state is dropped.

fn set_in_reasoning(&mut self, _in_reasoning: bool)

Override the parser’s initial reasoning state. When called with true, the parser starts in reasoning mode without waiting for the start token in the completion stream. Use this when the chat template already injected the start token (e.g., <think>) into the prompt, so it won’t appear in the model’s output.