o
ie @ s d dl mZ d dlmZ d dlmZ d dlmZ d dlm Z d dl
m Z d dl m
Z
d dlmZ e eZG d d
d
e
ZG d d d e Zd
S ) )Sequence)ChatCompletionRequest) DeltaMessage)ResponsesRequest) init_logger)ReasoningParser)BaseThinkingReasoningParser)
TokenizerLikec @ sj e Zd ZdZedefddZedefddZdeded ed
ee d ee d ee de
d
B fddZ d
S )MiniMaxM2ReasoningParserz
Reasoning parser for MiniMax M2 model.
MiniMax M2 models don't generate start token, only end
token. All content before is reasoning, content after is the
actual response.
returnc C dS )z(The token that starts reasoning content. selfr r `/home/ubuntu/vllm_env/lib/python3.10/site-packages/vllm/reasoning/minimax_m2_reasoning_parser.py start_token z$MiniMaxM2ReasoningParser.start_tokenc C r )z&The token that ends reasoning content.r r r r r end_token% r z"MiniMaxM2ReasoningParser.end_token
previous_text current_text
delta_textprevious_token_idscurrent_token_idsdelta_token_idsNc
C s t |dkr|d | jkrdS | j|v rt|dS | j|v rE|| j}|d| }||t | j d } t|r:|nd| rA| dS ddS t|dS )z
Extract reasoning content from a delta message for streaming.
MiniMax M2 models don't generate start token, so we assume
all content is reasoning until we encounter the end token.
r Ncontent) reasoningr )r )len end_token_idr findr )
r r r r r r r end_indexr r r r r extract_reasoning_streaming* s
z4MiniMaxM2ReasoningParser.extract_reasoning_streaming) __name__
__module__ __qualname____doc__propertystrr r r intr r$ r r r r r
s* r
c s e Zd ZdZdef fdd Zdee defddZ de
e de
e fd d
Z d e d e d
e dee dee dee de
dB fddZde deeB dee dB e dB f fddZ ZS )#MiniMaxM2AppendThinkReasoningParserz0
Reasoning parser for MiniMax M2 model.
tokenizerc s, t j|g|R i | | jd| _d S )Nr )super__init__vocabgetr! )r r- argskwargs __class__r r r/ W s z,MiniMaxM2AppendThinkReasoningParser.__init__ input_idsr c s | j t fddt|D S )Nc 3 s | ]}| kV qd S Nr ).0input_idr! r r ] s zGMiniMaxM2AppendThinkReasoningParser.is_reasoning_end..)r! anyreversedr r6 r r: r is_reasoning_end[ s z4MiniMaxM2AppendThinkReasoningParser.is_reasoning_endc C s |S r7 r r> r r r extract_content_ids_ s z7MiniMaxM2AppendThinkReasoningParser.extract_content_idsr r r r r r Nc C s t |dkr
d| }t|dS )Nr r
r )r r )r r r r r r r r r r r$ b s
z?MiniMaxM2AppendThinkReasoningParser.extract_reasoning_streaming model_outputrequestc C s d d| fS )Nr
r )r rA rB r r r extract_reasoningo s z5MiniMaxM2AppendThinkReasoningParser.extract_reasoning)r% r& r' r( r r/ r r+ boolr? listr@ r* r r$ r r tuplerC
__classcell__r r r4 r r, R s6
r, N)collections.abcr 0vllm.entrypoints.openai.chat_completion.protocolr 'vllm.entrypoints.openai.engine.protocolr *vllm.entrypoints.openai.responses.protocolr vllm.loggerr $vllm.reasoning.abs_reasoning_parsersr vllm.reasoning.basic_parsersr vllm.tokenizersr r% loggerr
r, r r r r s ;