o o i"@s$ddlZddlZddlZddlmZmZmZmZddlZddl m Z ddl m Z m Z ddlmZddlmZddlmZddlmZdd lmZdd lmZdd lmZmZer^dd lmZeeefZGd ddeZGdddeZddZ GdddeZe!dddgZ"GdddeZ#dS)N) TYPE_CHECKINGAny GeneratorUnion) DedentError)UnexpectedCharactersUnexpectedToken)Generate)Guide) RegexGuide)Write)create_states_mapping)grammars) PartialLarkPartialParserState) Tokenizerc@seZdZUdZeed<dS)r akBase definition of a generation guide. A generation guide defines the behavior of a finite-state machine that guides a text generation procedure. Unlike the DFAs built from regular expressions guides can also emit a `Write` instructions which tells the model that it can append a sequence of tokens (or token word) instead of generating it. initial_stateN)__name__ __module__ __qualname____doc__r__annotations__rr^/home/app/PaddleOCR-VL-test/.venv_paddleocr/lib/python3.10/site-packages/outlines/fsm/guide.pyr s  r c@sdeZdZdZdZdZdZdddZded e fd d Z ded ed efd dZ defddZ ddZ dS)StopAtEOSGuidez@Guide to generate tokens until the EOS token has been generated.r tokenizerrcCs|j|_|j|_dS)zzInitialize the generation guide. model The logit generator used to generate the next token. N) eos_token_id vocabularyvalues)selfrrrr__init__/szStopAtEOSGuide.__init__statereturncCs||r t|jgStdSN)is_final_stater rr r r"rrrget_next_instruction9s  z#StopAtEOSGuide.get_next_instructiontoken_idcCs ||jks ||jkr |jS|jSr$)r final_stater)r r"r(rrrget_next_state>szStopAtEOSGuide.get_next_statecCs ||jkSr$)r)r&rrrr%Ds zStopAtEOSGuide.is_final_statecCs|Sr$rr rrrcopyGszStopAtEOSGuide.copyN)rr)rrrrr)Z start_staterr!int Instructionr'r*r%r,rrrrr(s   rcOst||g|Ri|Sr$)uncached_create_states_mapping) regex_stringrargskwargsrrrcached_create_states_mappingKsr3cs*eZdZdZedeffdd ZZS)r zp Guide to generate text in the language of a regular expression. CoreRegexGuide with outlines cache r0c stj||fdti|S)NZ_create_states_mapping)super from_regexr3)clsr0rr2 __class__rrr5UszRegexGuide.from_regex)rrrr classmethodstrr5 __classcell__rrr7rr Os r CFGState parser_state prev_tokenc @seZdZdZdefddZdedefddZded e de e d d ffd d Z ded e defddZ ded e defddZdedefddZdedefddZdedefddZdddZd S)CFGGuidezNGuide to generate text that is in the language of a context-free Lark grammar. cfg_stringcCsNtd||_||_|jj|_t|dtjgd|_t |j ddd|_ dS)zg Construct the PartialLark parser and set the empty initial_state (PartialParserState) zOutlines' public *community-contributed* CFG structured generation is experimental. Please review https://dottxt-ai.github.io/outlines/latest/reference/generation/cfg#disclaimerZlalr)parserZ import_pathsNr=r>) warningswarnr@rrrrZ GRAMMAR_PATHrAr<parser)r r@rrrrr!js   zCFGGuide.__init__r"r#cCs\|jdurtt|jgSt|||jj }t |dkr'tt|St t|S)aWReturn the next instruction for guided generation. Current lazy approach: - For each token in the vocabulary - create a copy of the parsers state - add the tokens to the parsers input text - if valid, add token to returned tokens Further refinements are necessary for performant text processing. Parameters ---------- state The guides current PartialParserState, or None if complete Returns ------- A `Generate` instance that contains the model and the allowed token ids. Nr) r=r torchZtensorrlistiter_valid_token_idsrrrlenr )r r"Z valid_tokensrrrr's  zCFGGuide.get_next_instructioncandidate_token_idsNc csx|jdur |jVdS|D]+}||jkr||r|Vqz ||t||VWqttttt fy9YqwdS)a Iterate over the given token_ids and yield those that are valid for the current parser state. Parameters ---------- parser_state The current state of the parser, or None if complete. token_ids The list of token ids to check for validity. Yields ------ int Valid token ids. N) r=rcan_terminate_state_get_parser_state_token_appliedr- ValueErrorEOFErrorrrr)r r"rKr(rrrrIs,     zCFGGuide.iter_valid_token_idsr(cCs6|jdus ||jkr d}n||t|}t||dS)a Update the state of the guide. Decode the token_id, and calculate the new parser_state with the token applied. Parameters ---------- state The guides current PartialParserState, or None if complete token_id The id of the token that was just generated. Returns ------- The guides new PartialParserState NrC)r=rrMr-r<)r r"r(r=rrrr*s zCFGGuide.get_next_statecCst|j}|jdur|j|gd}n|j|jggd}|j|j|ggd}|t|d}|dkrrdecoderJrNlexerr"textrAZparse_from_state)r r"r(r=Z new_token_strZprev_token_strZcombined_token_strrrrrMs z(CFGGuide._get_parser_state_token_appliedcCs ||Sr$)rLr&rrrr%s zCFGGuide.is_final_statecCs:|jdurz t|jWdStyYdSwdS)z"Generation is allowed to terminateNFT)r=r,feed_eofrr&rrrrLs  zCFGGuide.can_terminate_statecCs |jdupt|jdhS)z1Generation must terminate, no legal continuationsNz$END)r=setZacceptsissubsetr&rrrmust_terminate_state szCFGGuide.must_terminate_statecCst|j|jS)zCreate a copy of the Guide.)r?r@rr+rrrr,sz CFGGuide.copy)r#r?)rrrrr:r!r<r.r'rHrr-rIr*rrMboolr%rLrVr,rrrrr?gs.   ' ! r?)$ collectionsr,rDtypingrrrrrGZ lark.indenterrZ lark.lexerrrZoutlines_core.fsm.guider r Z CoreGuider ZCoreRegexGuider r r/ZoutlinesrZoutlines.fsm.parsingrrZoutlines.models.tokenizerrr.rr3 namedtupler<r?rrrrs.          #