o iY7 @s$ddlmZmZddlmZmZmZddlmZddl m Z ddl m Z m Z mZddlmZddlmZddlmZerAdd lZned ed ZeeZed eed Zd ddeeefd Bdeeefd BdeedfdeeeffddZ eddGdddZ!eddGdddZ"d S)) dataclassfield) TYPE_CHECKINGAnyTypeVar)ChatTemplateContentFormatOption)VLLMValidationError) EmbedsPrompt TextPrompt TokensPrompt) init_logger) TokenizerLike) LazyLoaderNtorch_Sz torch.Tensor)Nauto unset_valuesdefaults overridesr.returncs2|duri}|dur i}|fdd|DBS)Ncsi|] \}}|vr||qSr).0kvrrK/home/ubuntu/vllm_env/lib/python3.10/site-packages/vllm/renderers/params.py $sz merge_kwargs..)items)rrrrrr merge_kwargss rT)frozenc@s~eZdZUdZdZedBed< dZeed< e e dZ e ee fed< de ee fdBfd d Z d e ee ffd d ZdS) ChatParamsz4Configuration to control how to parse chat messages.N chat_templaterchat_template_content_format)default_factorychat_template_kwargsdefault_chat_template_kwargscCs"|s|St|j|jt||jdS)N)r!r"r$)r r!r"rr$)selfr%rrr with_defaults4szChatParams.with_defaultsrcCst|jt|jddS)z9The arguments to pass to `tokenizer.apply_chat_template`.F)r! return_dict)rr$dictr!r&rrrget_apply_chat_template_kwargsAs z)ChatParams.get_apply_chat_template_kwargs)__name__ __module__ __qualname____doc__r!str__annotations__r"rrr)r$rr'r+rrrrr 's   r c@seZdZUdZedBed< dZeed< dZedBed< dZedBed< dZ e ed < d Z e ed < dZ e ed < dZ eed < dZeed< dZeed< ededBfddZd0ddZdeeefdBfddZdeeeffddZdedBdedefddZdedBdedefddZdedBdedefd d!ZdedBd"edefd#d$ZdedBd%edefd&d'ZdedBd%edefd(d)Z dedBd%edefd*d+Z!dedBd%edefd,d-Z"dedBd"e#e$Bde#e$Bfd.d/Z%dS)1TokenizeParamsz3Configuration to control how prompts are tokenized.Nmax_total_tokensrmax_output_tokenspad_prompt_tokenstruncate_prompt_tokensF do_lower_caseTadd_special_tokensneeds_detokenizationmax_total_tokens_parammax_output_tokens_paramtruncate_prompt_tokens_paramrcCs|jdurdS|j|jS)z'Maximum allowed number of input tokens.N)r3r4r*rrrmax_input_tokens|s  zTokenizeParams.max_input_tokensc Cs|j}|j}|j}|j}|dur.|dur.||kr.t|jd|d|jd|d|j|d|durT|durV||krXt|jd|d|jd|jd|d |j|ddSdSdS) N=zcannot be greater than z=max_total_tokens=z%. Please request fewer output tokens. parametervaluez cannot be greater than z - z = z+. Please request a smaller truncation size.)r3r4r=r6rr;r:r<)r&r3r4r=r6rrr __post_init__sB zTokenizeParams.__post_init__tokenization_kwargsc Cs |duri}|d|j}|d|j}|d|j}|d|j}|d|j}|d|j}|dd}rJ|dkr?|}n |dvrFd}n||d<|d d} rd| d vrY|}n | d vr`d}n| |d <|rltd ||j } t | | dusy|dur{d n| ||||||dS)N max_lengthr5r6r7r8r9padding)F do_not_pad truncation)T longest_first)Fdo_not_truncatez_The following tokenization arguments are not supported by vLLM Renderer and will be ignored: %sr)r3r4r5r6r7r8r9) popr=r5r6r7r8r9loggerwarningr3r2) r&rCrDr5r6r7r8r9rErGr3rrr with_kwargssZzTokenizeParams.with_kwargscCsN|j}|dur|dkr|j}n|dur|jdur|jd}t|du||jdS)z,The arguments to pass to `tokenizer.encode`.Nr)rGrDr8)r6r=r)r8)r&rDrrrget_encode_kwargss z TokenizeParams.get_encode_kwargs tokenizertextc Csx|j}|dur |S|jdur:|dur:||j}t||kr:tdt|d|jd|jd|d|d dt|d |S) z0Apply length checks to prompt text if necessary.N You passed z input characters and requested < output tokens. However, the model's context length is only 0 tokens, resulting in a maximum input length of z tokens (at most z; characters). Please reduce the length of the input prompt. input_textr?)r=r6max_chars_per_tokenlenrr4r3)r&rPrQr=max_input_charsrrr_text_len_checks(    zTokenizeParams._text_len_checkcCs|jr|S|S)z,Apply lowercase to prompt text if necessary.)r7lower)r&rPrQrrr_text_lowercase szTokenizeParams._text_lowercasecCs |j|jfD]}|||}q|S)z$Apply all validators to prompt text.)rYr[)r&rPrQ validatorrrr_validate_texts  zTokenizeParams._validate_textpromptcCs|||d|d<|S)z Ensure that the prompt meets the requirements set out by this config. If that is not possible, raise a `VLLMValidationError`. This method is run before tokenization occurs. r^)r]r&rPr^rrrapply_pre_tokenizations z%TokenizeParams.apply_pre_tokenizationtokenscCsn|j}|dur|dkr|j}|dus|t|kr|S|dur"tdt|ts+td||jg|t|S)z,Apply padding to prompt tokens if necessary.Nrz1Cannot pad tokens when `skip_tokenizer_init=True`z&Cannot pad tokens for embedding inputs)r5r=rW ValueError isinstancelist pad_token_id)r&rPra pad_lengthrrr_token_padding*s zTokenizeParams._token_paddingcCsr|j}|dur|dkr|j}|dus|t|kr|S|dkr$|ddSt|dddkr3|| dS|d|S)z/Apply truncation to prompt tokens if necessary.Nrtruncation_sideleft)r6r=rWgetattr)r&rPrarDrrr_token_truncation:s  z TokenizeParams._token_truncationc CsV|j}|dur |St||kr)tdt|d|jd|jd|d dt|d|S) z2Apply length checks to prompt tokens if necessary.NrRz input tokens and requested rSrTz6 tokens. Please reduce the length of the input prompt. input_tokensr?)r=rWrr4r3)r&rPrar=rrr_token_len_checkJs    zTokenizeParams._token_len_checkcCs$|j|j|jfD]}|||}q|S)z)Apply all validators to a token sequence.)rgrkrm)r&rPrar\rrr_validate_tokens^s  zTokenizeParams._validate_tokenscCs<d|vr|||d|d<d|vr|||d|d<|S)z Ensure that the prompt meets the requirements set out by this config. If that is not possible, raise a `VLLMValidationError`. This method is run after tokenization occurs. prompt_token_ids prompt_embeds)rnr_rrrapply_post_tokenizationis z&TokenizeParams.apply_post_tokenization)rN)&r,r-r.r/intr1r4r5r6r7boolr8r9r:r0r;r<propertyr=rBr)rrMrOr rYr[r]r r`rrgrkrmrnr r rqrrrrr2Is`          "=   r2)# dataclassesrrtypingrrrvllm.entrypoints.chat_utilsrvllm.exceptionsr vllm.inputsr r r vllm.loggerr vllm.tokenizersr vllm.utils.import_utilsrrglobalsr,rKrdrrrr)r0tupleobjectrr r2rrrrs6         !