`i . ddlZddlZddlmZddlmZn#e$r edwxYwddlmZm Z m Z m Z ddl m Z ddlmZmZmZddlZGdd Zdd eejefd ed edzde fdZ ddeejee fdededefdZddgZdS)N)MistralTokenizer)PreTrainedTokenizerBasez@vllm is not installed. Please install it with "pip install vllm")CharacterLevelParser TokenEnforcerFormatEnforcerAnalyzerTokenEnforcerTokenizerData)#build_token_enforcer_tokenizer_data)ListOptionalUnioncNeZdZdefdZdeedejdejfdZ dS)VLLMLogitsProcessortoken_enforcercR||_|rt|nd|_d|_dS)N)rranalyzermask)selfranalyzes v/home/jaya/work/projects/VOICE-AGENT/VIET/agent-env/lib/python3.11/site-packages/lmformatenforcer/integrations/vllm.py__init__zVLLMLogitsProcessor.__init__s/,BIS.~>>>t ,0 input_idsscoresreturnc||}|jr-|j|||j|j}|j&|jtj n%tj |tj |_d|j|<||jz}|S)Nr) rreport_raw_logitstolistrget_allowed_tokensallowed_tokensrfill_mathinftorch full_like)rrrtoken_sequencers r__call__zVLLMLogitsProcessor.__call__s" = M M + +NFMMOO L L L,??OO^ 9 IOOTXI & & & & ::DI$% .!$)# rN) __name__ __module__ __qualname__rrr intr#Tensorr&rrrrs`1}1111 $s) U\ el      rrF tokenizer use_bitmask vocab_sizerc`|;t|dr+|j}t|dr|}t |t rt|||St|dr|j}t|||S)N llm_engine get_tokenizerr-) hasattrr1get_model_configget_vocab_sizer2 isinstancerr r-)r-r.r/s r(build_vllm_token_enforcer_tokenizer_datar7#s 9l + + R"->>@@OOQQJy/**.++-- )-..W29k:VVVy+&&(' .y+z R RRrllmcharacter_level_parserrct|tst|}t||}t ||S)zBuild the logits processor function that llama.cpp will use to filter the tokens generated by the model. The result can be passed in the logits_processor list that is sent to the call or generate() method of llama.cpp models.)r6rr7rr)r8r9rrs rbuild_vllm_logits_processorr;1sD c5 6 6<6s;;"3(>??N ~w 7 77rr;r7)FN)F)r#vllm!vllm.transformers_utils.tokenizerr transformersr ImportErrorlmformatenforcerrrrr*lmformatenforcer.integrations.transformersr typingr r r r!rLLMboolr*r7r;__all__r,rrrFsZLLLKKKBBBBBB4444444ZZZ +X Y YYZttttttttttttZZZZZZ(((((((((( * S SdhH_>_8` Sos SJMPTJT S`z S S S S /488U485LNh+h%i88L8)-88K8888 )*T Us(