o ̳i( @sddlmZmZmZmZmZmZddlmZddl m Z ddl m Z ddl mZddlmZmZddd d d d d ddddd ZGdddeeZdS))AnyDictListMappingOptionalTuple)Message)PromptTemplate)truncate) Transform)ModelTokenizerSentencePieceBaseTokenizer}i}i}i}i}i}i}i}i}i }i }) <|endoftext|> <|assistant|>z<|placeholder1|>z<|placeholder2|>z<|placeholder3|>z<|placeholder4|> <|system|><|end|>z<|placeholder5|>z<|placeholder6|><|user|>c @seZdZdZ   d#dedeeeefdeedeefddZ e d d Z e d d Z d$dede de de deef ddZd%deede defddZddddeede de deeeee ffddZ d&deeefd e deeeffd!d"ZdS)'Phi3MiniTokenizera SentencePiece tokenizer configured with Phi3 Mini's special tokens. Args: path (str): Path to pretrained tokenizer file. special_tokens (Optional[Dict[str, int]]): mapping containing special text tokens and their registered token IDs. If left as None, this will be set to the canonical Phi3 special tokens. max_seq_len (Optional[int]): A max sequence length to truncate tokens to. Default: None prompt_template (Optional[PromptTemplate]): template used to format the messages based on their role. This is used to add structured text around the actual messages. The structured text is used in three scenarios: - Task-specific templates to gear models for a particular task that it will expect after training - Model-specific templates that are required whenever the model is prompted, such as the [INST] tags in Llama2 and in Mistral - Community standardized templates, such as :class:`~torchtune.data.ChatMLTemplate` The extra text will still get tokenized as normal text, not as special tokens. Default is None. Examples: >>> tokenizer = Phi3MiniTokenizer("/path/to/spm_model") >>> tokenized_text = tokenizer.encode("Hello world!", add_bos=True, add_eos=True) >>> print(tokenized_text) [1, 31587, 29644, 102, 2] Npathspecial_tokens max_seq_lenprompt_templatecCsNt||_|dur |nt|_|jd|_|jd|_|jg|_||_||_dS)Nr) r _spm_modelPHI3_SPECIAL_TOKENSreos_idpad_id stop_tokensrr)selfrrrrrT/home/ubuntu/.local/lib/python3.10/site-packages/torchtune/models/phi3/_tokenizer.py__init__=s     zPhi3MiniTokenizer.__init__cC|jjSN)r vocab_sizerrrr r$UzPhi3MiniTokenizer.vocab_sizecCr"r#)rbos_idr%rrr r'Yr&zPhi3MiniTokenizer.bos_idTFtextadd_bosadd_eostrim_leading_whitespacereturncCs|jj||||dS)Nr)r*r+)rencode)rr(r)r*r+rrr r.]s zPhi3MiniTokenizer.encodeidsskip_special_tokenscCs:g}|D]}|r|dkr|dkrq||q|j|S)a0Decode token IDs to strings. Args: ids (List[int]): The input token IDs to be decoded. skip_special_tokens (bool): Whether to show or skip special tokens in the decoded string. Default is True. Returns: str: The decoded text. ri@})appendrdecode)rr/r0ids_for_decodetoken_idrrr r2ks    zPhi3MiniTokenizer.decode)r*ignore_system_promptmessagesr5c Cs@|jdur ||n|}d}d}g}g}|jdddd} |D]} |r(| jdkr(q|r6||j|| j| jdkrJ||jd|| jn7| jd kr`||jd d}|| jn!| jdkrt||jd || jn td | jd | jd| | | | jgt | g} | jD]!} | ddkr| |j| d ddddd} qt d| d| |jdg| } | | | | jgt | |r|r||j || jd}d}nd}|jrt ||jkrnq|jrt ||jkrt||j|r |j nd}t||j|r| jnd}||fS)aTokenize a list of messages one at a time then concatenate them, returning a list of tokens and a list of masks. Example: >>> tokenizer = Phi3MiniTokenizer(tokenizer_path, max_seq_len) >>> messages = [ Message(role="system", content="system message\n", masked=True), Message(role="user", content="user prompt\n", masked=True), Message(role="assistant", content="assistant response\n"), ] >>> # tokenize_messages encodes messages separately and concats >>> tokenizer.tokenize_messages(messages)[0] [1, 1788, 2643, 13, 1792, 9508, 13, 465, 22137, 2933, 2] >>> # Same result as encoding the full string in one go >>> tokenizer.encode(''.join([message.content for message in messages])) [1, 1788, 2643, 13, 1792, 9508, 13, 465, 22137, 2933, 2] Args: messages (List[Message]): A list of messages, each containing role, content, and masked attributes. add_eos (bool): Whether to append EOS after assistant message, default to False ignore_system_prompt (bool): Whether to ignore system prompt, defaults to False. Raises: ValueError: If the role is not "user", "assistant", or "system". RuntimeError: If ``message["type"] != "text``. Returns: Tuple[List[int], List[bool]]: The tokenized messages NTF )r)r*systemuserr assistantrrzUnknown role 'z' for message: ''typer(content r-z"Unsupported message content type: r)rr.roler1r'maskedr ValueErrorr=extendlenrstrip RuntimeErrorrrr ) rr6r*r5templated_messages start_of_turn end_of_turntokenized_messagesmasknew_line_token_idmessagetokensitemrrr tokenize_messagessx *               z#Phi3MiniTokenizer.tokenize_messagessample inferencecCs,|d}||\}}||d<||d<|S)a% Apply ``tokenize_messages`` to the "messages" field in the sample. Args: sample (Mapping[str, Any]): A sample with a "messages" field containing a List[Message] to tokenize inference (bool): Whether the template is being used for inference or not. Returns: Mapping[str, Any]: The sample with added "tokens" and "mask" fields and the "messages" field removed. inference (bool): Whether the template is being used for inference or not. r6rMrJ)poprO)rrPrQr6rMrJrrr __call__s zPhi3MiniTokenizer.__call__)NNN)TTF)T)F)__name__ __module__ __qualname____doc__strrrintr r!propertyr$r'boolrr.r2rrrOrrrSrrrr r!sh       rN)typingrrrrrrtorchtune.data._messagesr torchtune.data._prompt_templatesr torchtune.data._utilsr torchtune.modules.transformsr 'torchtune.modules.transforms.tokenizersr r rrrrrr s&