o
-i
@ s| d dl mZ d dlmZ d dlmZ d dlmZ d dlm Z d dl
m Z d dl m
Z
d dlmZ eeZG d d
d
e Zd S ) )Sequence)ChatCompletionRequest) DeltaMessage) init_logger)ReasoningParser)DeepSeekR1ReasoningParser)IdentityReasoningParser)
TokenizerLikec s e Zd ZdZdef fdd Zdee defddZ dee d ee defd
d Z
de e de e fd d
Z de
dedee
dB e
dB f fddZde
de
de
dee dee dee dedB fddZ ZS )Holo2ReasoningParsera
Reasoning parser for the Holo2 models which are based on Qwen3.
The Holo2 model uses ... tokens to denote reasoning text but
is part of the chat template. This parser extracts the reasoning content until
in the model's output.
The model provides a switch to enable or disable reasoning
output via the 'thinking=False' parameter.
Chat template args:
- thinking: Whether to enable reasoning output (default: True)
Parsing rules on model output:
- thinking == False
-> Model output is treated as purely the content |content|
- thinking == True
-> Model output is |reasoning_content||content|
tokenizerc s t j|g|R i | |di pi }t|dd}t|dd}|o(|}|r9t|g|R i || _d S t|g|R i || _d S )Nchat_template_kwargsthinkingTenable_thinking)super__init__getboolr _parserr )selfr argskwargs chat_kwargsr
r __class__ b/home/ubuntu/veenaModal/venv/lib/python3.10/site-packages/vllm/reasoning/holo2_reasoning_parser.pyr + s z Holo2ReasoningParser.__init__ input_idsreturnc C | j |S N)r is_reasoning_endr r r r r r < z%Holo2ReasoningParser.is_reasoning_end delta_idsc C | j ||S r )r is_reasoning_end_streaming)r r r# r r r r% ? z/Holo2ReasoningParser.is_reasoning_end_streamingc C r r )r extract_content_idsr! r r r r' D r" z(Holo2ReasoningParser.extract_content_ids model_outputrequestNc C r$ r )r extract_reasoning)r r( r) r r r r* G r& z&Holo2ReasoningParser.extract_reasoning
previous_text current_text
delta_textprevious_token_idscurrent_token_idsdelta_token_idsc C s | j ||||||S r )r extract_reasoning_streaming)r r+ r, r- r. r/ r0 r r r r1 L s z0Holo2ReasoningParser.extract_reasoning_streaming)__name__
__module__ __qualname____doc__r r r intr r r% listr' strr tupler* r r1
__classcell__r r r r r
sD
r
N)collections.abcr 0vllm.entrypoints.openai.chat_completion.protocolr 'vllm.entrypoints.openai.engine.protocolr vllm.loggerr vllm.reasoningr +vllm.reasoning.deepseek_r1_reasoning_parserr (vllm.reasoning.identity_reasoning_parserr vllm.tokenizersr r2 loggerr
r r r r s