Pi&fddlmZddlmZmZmZmZmZddlm Z ddl m Z m Z ee ee ee e fffZGddeZGddeZGd d eZeed d i Zde_eed di Zde_eed di Zde_dedefdZdS))partial)DictListProtocolTupleUnion)_get_component_from_path)MessageRolecneZdZUdZeeeeeffed< d de e de de e fdZ dS) PromptTemplateInterfacez Interface for prompt templates. Each prompt template can include structured text for system, user, and assistant roles that are prepended or appended to the message content. templateFmessages inferencereturncdS)a Format each role's message(s) according to the prompt template Args: messages (List[Message]): a single conversation, structured as a list of :class:`~torchtune.data.Message` objects inference (bool): Whether the template is being used for inference or not. Returns: The formatted list of messages N)selfrrs t/home/jaya/work/projects/VOICE-AGENT/VIET/agent-env/lib/python3.11/site-packages/torchtune/data/_prompt_templates.py__call__z PromptTemplateInterface.__call__s NF) __name__ __module__ __qualname____doc__rr rstr__annotations__rr boolrrrrr r s4sCx())))    w-   g       rr cneZdZdZdeeeeefffdZ d de e de de e fdZ d S) PromptTemplateaI Quickly define a custom prompt template by passing in a dictionary mapping role to the prepend and append tags. For example, to achieve the following prompt template:: System: {content}\n User: {content}\n Assistant: {content}\n Tool: {content}\n You need to pass in a tuple for each role, where ``PREPEND_TAG`` is the string added before the text content and ``APPEND_TAG`` is the string added after:: template = {role: (PREPEND_TAG, APPEND_TAG)} Thus, the template would be defined as follows:: template = { "system": ("System: ", "\n"), "user": ("User: ", "\n"), "assistant": ("Assistant: ", "\n"), "ipython": ("Tool: ", "\n"), } Once instantiated, you must call the prompt template on a list of messages. It will return the same list of messages updated with the template. Note: Any tags prepended/appended to the assistant message will be included in the loss calculation. All other prepend/append tags for other roles (system, user, ipython) are, in most cases, not included in loss. Consider using the append tags for user messages for tags that need to come before the assistant message but should not be included in loss. For more custom masking and prompt templating, you can create your own class based off the :class:`~torchtune.data.PromptTemplate` interface. Args: template (Dict[Role, Tuple[str, str]]): a dictionary mapping role to the prepend and append tags rc||_dS)Nr)rrs r__init__zPromptTemplate.__init__Xs! rFrrrc g}|D]}|j}|j|jvr|j|jd}|j|jd}|j}t|trt |dkr d|dg|z}t|trt |dkr |d|dgz}|t|j||j|j |j |S)a Format each role's message(s) according to the prompt template by prepending and appending the defined tags. Args: messages (List[Message]): list of messages to apply the template to inference (bool): Whether the template is being used for inference or not. Returns: List[Message]: The formatted list of messages rtexttypecontentroler*maskedipythoneot) r*r,r isinstancerlenappendr r-r.r/)rrrformatted_dialoguemessager* prepend_tag append_tags rrzPromptTemplate.__call__^s   GoG|t},,"mGL9!< !]7<8; !/k3//SC 4D4Dq4H4H(.;GGH7RGj#..R3z??Q3F3F%&Z)P)P(QQG  % % #">#O     "!rNr) rrrrrr rrr$rr rrrrrr!r!.s''R!tU38_,-!!!!:?$"$"W $"26$" g$"$"$"$"$"$"rr!cPeZdZdZdddddZ d deed ed eefd Zd S)ChatMLTemplateuE OpenAI's `Chat Markup Language `_ used by their chat models. It is the default chat template used by Hugging Face models. .. code-block:: text <|im_start|>system Provide some context and/or instructions to the model.<|im_end|> <|im_start|>user The user’s message goes here<|im_end|> <|im_start|>assistant The assistant’s response goes here<|im_end|> )z<|im_start|>system <|im_end|> )z<|im_start|>user r9)z<|im_start|>assistant r9)r:)systemuser assistantr.Frrrc g}t|D]Q\}}|j|jd}|j|jd}|jdkrl|t|dz krVt|jdkr>|j}t |tr!t|dkrd|dg|jz}ni|j}t |trt|dkr d|dg|z}t |trt|dkr |d|dgz}|t|j||j |j |j S|S)an Format user, assistant, and system messages with appropriate tags. Args: messages (List[Message]): a single conversation, structured as a list of `Message` objects inference (bool): Whether the template is being used for inference or not. Returns: The formatted list of messages rr&r=r'r(r+) enumeraterr,r1 text_contentr*r0rr2r r-r.r/) rrrr3indexr4r5r6r*s rrzChatMLTemplate.__call__s  '11! ! NE7- 5a8Kw|4Q7J  ++S]]Q...,--22!/k3//(C 4D4Dq4H4H!'K@@(G"/k3//SC 4D4Dq4H4H(.;GGH7RGj#..R3z??Q3F3F%&Z)P)P(QQG  % % #">#O     "!rNr) rrrrrrr rrrrrr8r8sz&;6@ H 3"3"w-3"3" g 3"3"3"3"3"3"rr8r<)z"Correct this to standard English: z --- Corrected: r#z A prompt template for grammar error correction tasks:: Correct this to standard English: {user_message} --- Corrected: {assistant_message} Please see :class:`~torchtune.data.PromptTemplate` for full API arguments. )zSummarize this dialogue: z --- Summary: z A prompt template for summarization tasks:: Summarize this dialogue: {user_message} --- Summary: {assistant_message} Please see :class:`~torchtune.data.PromptTemplate` for full API arguments. )z Question: z Answer: z A prompt template for question answering tasks:: Question: {user_message} Answer: {assistant_message} Please see :class:`~torchtune.data.PromptTemplate` for full API arguments. prompt_templaterct|trt|St|trt |St dt |)aZ Retrieve prompt template from import dotpath or create a custom one with provided template dictionary. Args: prompt_template (_TemplateType): optional specified prompt template. If a string, it is assumed to be the dotpath of a :class:`~torchtune.data.PromptTemplateInterface` class. If a dictionary, it is assumed to be a custom prompt template mapping role to the prepend/append tags. Returns: PromptTemplateInterface: the specified prompt template Raises: ValueError: If a string or dictionary is not passed in zQPrompt template must be a dotpath string or dictionary with custom template, got )r0rr dictr! ValueErrorr))rBs r_get_prompt_templaterFst&/3'' 8'88::: OT * * o... w`det`u`u w w   rN) functoolsrtypingrrrrrtorchtune.config._utilsr torchtune.data._messagesr r r _TemplateTyper r!r8GrammarErrorCorrectionTemplaterSummarizeTemplateQuestionAnswerTemplaterFrrrrOs 55555555555555<<<<<<22222222c4eCHo 5667      h