import gradio as gr import json from difflib import Differ, unified_diff from itertools import groupby from gradio_huggingfacehub_search import HuggingfaceHubSearch from huggingface_hub import HfApi, CommitOperationAdd from transformers import PreTrainedTokenizerBase from enum import StrEnum from copy import deepcopy hfapi = HfApi() class ModelFiles(StrEnum): CHAT_TEMPLATE_JSON = "chat_template.json" TOKENIZER_CHAT_TEMPLATE = "tokenizer_chat_template.jinja" TOKENIZER_CONFIG = "tokenizer_config.json" TOKENIZER_INVERSE_TEMPLATE = "inverse_template.jinja" example_labels = [ "Single user message", "Single user message with system prompt", "Longer conversation", "Tool call", "Tool call with response", "Tool call with multiple responses", "Tool call with complex tool definition", "RAG call", "Multimodal user message", ] example_values = [ [ "{}", """[ { "role": "user", "content": "What is the capital of Norway?" } ]""", ], [ "{}", """[ { "role": "system", "content": "You are a somewhat helpful AI." }, { "role": "user", "content": "What is the capital of Norway?" } ]""", ], [ "{}", """[ { "role": "user", "content": "What is the capital of Norway?" }, { "role": "assistant", "content": "Oslo is the capital of Norway." }, { "role": "user", "content": "What is the world famous sculpture park there called?" }, { "role": "assistant", "content": "The world famous sculpture park in Oslo is called Vigelandsparken." }, { "role": "user", "content": "What is the most famous sculpture in the park?" } ]""", ], [ """{ "tools": [ { "type": "function", "function": { "name": "get_current_weather", "description": "Get the current weather in a given location", "parameters": { "type": "object", "properties": { "location": { "type": "string", "description": "The city and state, e.g. San Francisco, CA" }, "unit": { "type": "string", "enum": [ "celsius", "fahrenheit" ] } }, "required": [ "location" ] } } } ] }""", """[ { "role": "user", "content": "What's the weather like in Oslo?" } ]""", ], [ """{ "tools": [ { "type": "function", "function": { "name": "get_current_weather", "description": "Get the current weather in a given location", "parameters": { "type": "object", "properties": { "location": { "type": "string", "description": "The city and state, e.g. San Francisco, CA" }, "unit": { "type": "string", "enum": [ "celsius", "fahrenheit" ] } }, "required": [ "location" ] } } } ] }""", """[ { "role": "user", "content": "What's the weather like in Oslo?" }, { "role": "assistant", "content": null, "tool_calls": [ { "id": "toolcall1", "type": "function", "function": { "name": "get_current_weather", "arguments": { "location": "Oslo, Norway", "unit": "celsius" } } } ] }, { "role": "tool", "content": "20", "tool_call_id": "toolcall1" } ]""", ], [ """{ "tools": [ { "type": "function", "function": { "name": "get_current_weather", "description": "Get the current weather in a given location", "parameters": { "type": "object", "properties": { "location": { "type": "string", "description": "The city and state, e.g. San Francisco, CA" }, "unit": { "type": "string", "enum": [ "celsius", "fahrenheit" ] } }, "required": [ "location" ] } } } ] }""", """[ { "role": "user", "content": "What's the weather like in Oslo and Stockholm?" }, { "role": "assistant", "content": null, "tool_calls": [ { "id": "toolcall1", "type": "function", "function": { "name": "get_current_weather", "arguments": { "location": "Oslo, Norway", "unit": "celsius" } } }, { "id": "toolcall2", "type": "function", "function": { "name": "get_current_weather", "arguments": { "location": "Stockholm, Sweden", "unit": "celsius" } } } ] }, { "role": "tool", "content": "20", "tool_call_id": "toolcall1" }, { "role": "tool", "content": "22", "tool_call_id": "toolcall2" } ]""", ], [ """{ "tools": [ { "type": "function", "function": { "name": "create_user", "description": "creates a user", "parameters": { "type": "object", "properties": { "user": { "title": "User", "type": "object", "properties": { "user_id": { "title": "User Id", "description": "The unique identifier for a user", "default": 0, "type": "integer" }, "name": { "title": "Name", "description": "The name of the user", "type": "string" }, "birthday": { "type": "string", "description": "The birthday of the user, e.g. 2022-01-01", "pattern": "^([1-9] |1[0-9]| 2[0-9]|3[0-1])(.|-)([1-9] |1[0-2])(.|-|)20[0-9][0-9]$" }, "email": { "title": "Email", "description": "The email address of the user", "type": "string" }, "friends": { "title": "Friends", "description": "List of friends of the user", "type": "array", "items": {"type": "string"} } }, "required": ["name", "email"] } }, "required": ["user"], "definitions": {} } } } ] }""", """[ { "role": "user", "content": "Create a user for Test User (test@user.org), born January 1st 2000, with some random friends." } ]""", ], [ """{ "documents": [ { "title": "Much ado about nothing", "content": "Dolor sit amet..." }, { "title": "Less ado about something", "content": "Lorem ipsum..." } ] }""", """[ { "role": "user", "content": "Write a brief summary of the following documents." } ]""", ], [ "{}", """[ { "role": "user", "content": [ { "type": "text", "content": "Can this animal" }, { "type": "image" }, { "type": "text", "content": "make this sound" }, { "type": "audio" }, { "type": "text", "content": "while moving like this?" }, { "type": "video" } ] } ]""" ] ] pr_description_default = "### Changes\n* \n\n**Updated using [Chat Template Editor](https://huggingface.co/spaces/CISCai/chat-template-editor)**" class TokenizerConfig(): def __init__(self, tokenizer_config: dict): self._data = deepcopy(tokenizer_config) self.chat_template = self._data.get("chat_template") @property def chat_template(self) -> str | list | None: templates = [ { "name": k, "template": v, } for k, v in self.chat_templates.items() if v ] if not templates: return None elif len(templates) == 1 and templates[0]["name"] == "default": return templates[0]["template"] else: return templates @chat_template.setter def chat_template(self, value: str | list | None): if not value: self.chat_templates.clear() elif isinstance(value, str): self.chat_templates = { "default": value, } else: self.chat_templates = { t["name"]: t["template"] for t in value } # @property # def inverse_template(self) -> str | None: # return self._data.get("inverse_template") # @inverse_template.setter # def inverse_template(self, value: str | None): # if value: # self._data["inverse_template"] = value # elif "inverse_template" in self._data: # del self._data["inverse_template"] def json(self, indent: int | str | None = 4) -> str: self._data["chat_template"] = self.chat_template return json.dumps(self._data, ensure_ascii = False, indent = indent) def get_json_indent( json: str, ) -> int | str | None: nonl = json.replace('\r', '').replace('\n', '') start = nonl.find("{") first = nonl.find('"') return "\t" if start >= 0 and nonl[start + 1] == "\t" else None if first == json.find('"') else min(max(first - start - 1, 2), 4) def character_diff( diff_title: str | None, str_original: str, str_updated: str, ): d = Differ() title = [] if diff_title is None else [ (f"\n@@ {diff_title} @@\n", "@") ] diffs = [ ("".join(map(lambda x: x[2:].replace("\t", "\u21e5").replace("\r", "\u240d\r").replace("\n", "\u240a\n") if x[0] != " " else x[2:], tokens)), group if group != " " else None) # .replace(" ", "\u2423") for group, tokens in groupby(d.compare(str_updated, str_original), lambda x: x[0]) ] return title + ([("No changes", "?")] if len(diffs) == 1 and diffs[0][1] is None else diffs) with gr.Blocks( ) as blocks: with gr.Row( equal_height = True, ): hf_search = HuggingfaceHubSearch( label = "Search Huggingface Hub", placeholder = "Search for models on Huggingface", search_type = "model", sumbit_on_select = True, scale = 2, ) hf_branch = gr.Dropdown( None, label = "Branch", scale = 1, ) gr.LoginButton( "Sign in for write access or gated/private repos", scale = 1, ) gr.Markdown( """# Chat Template Editor Any model repository with chat template(s) is supported (including GGUFs), however do note that all the model info is extracted using the Hugging Face API. For GGUFs in particular this means that the chat template may deviate from the actual content in any given GGUF file as only the default template from an arbitrary GGUF file is returned. If you sign in and grant this editor write access you will get the option to create a pull request of your changes (provided you have access to the repository). You can freely edit and test GGUF chat template(s) (and are encouraged to do so), but you cannot commit any changes, it is recommended to use the [GGUF Editor](https://huggingface.co/spaces/CISCai/gguf-editor) to save the final result to a GGUF. """, ) with gr.Accordion("Commit Changes", open = False, visible = False) as pr_group: with gr.Tabs() as pr_tabs: with gr.Tab("Edit", id = "edit") as pr_edit_tab: pr_title = gr.Textbox( placeholder = "Title", show_label = False, max_lines = 1, interactive = True, ) pr_description = gr.Code( label = "Description", language = "markdown", lines = 10, max_lines = 10, interactive = True, ) with gr.Tab("Preview (with diffs)", id = "preview") as pr_preview_tab: pr_preview_title = gr.Textbox( show_label = False, max_lines = 1, interactive = False, ) pr_preview_description = gr.Markdown( label = "Description", height = "13rem", container = True, ) pr_preview_diff = gr.HighlightedText( label = "Diff", combine_adjacent = True, color_map = { "+": "red", "-": "green", "@": "blue", "?": "blue" }, interactive = False, show_legend = False, show_inline_category = False, ) pr_submit = gr.Button( "Create Pull Request", variant = "huggingface", interactive = False, ) pr_submit.click( lambda: gr.Button( interactive = False, ), outputs = [ pr_submit, ], show_api = False, ) with gr.Tabs() as template_tabs: with gr.Tab("Edit", id = "edit") as edit_tab: with gr.Accordion("Template Input", open = False): chat_settings = gr.Code( label = "Template Settings (kwargs)", language = "json", interactive = True, render = False, ) chat_messages = gr.Code( label = "Template Messages", language = "json", interactive = True, render = False, ) example_input = gr.Examples( examples = example_values, example_labels = example_labels, inputs = [ chat_settings, chat_messages, ], ) chat_settings.render() chat_messages.render() chat_template = gr.Code( label = "Chat Template (default)", language = "jinja2", interactive = True, ) with gr.Accordion("Additional Templates", open = False): inverse_template = gr.Code( label = "Inverse Template", language = "jinja2", interactive = True, visible = False, ) chat_template_tool_use = gr.Code( label = "Chat Template (tool_use)", language = "jinja2", interactive = True, ) chat_template_rag = gr.Code( label = "Chat Template (rag)", language = "jinja2", interactive = True, ) with gr.Tab("Render", id = "render") as render_tab: rendered_chat_template = gr.Textbox( label = "Chat Prompt (default)", interactive = False, lines = 20, show_copy_button = True, ) with gr.Accordion("Additional Output", open = False): rendered_inverse_template = gr.Code( label = "Inverse Chat Messages", language = "json", interactive = False, visible = False, ) rendered_chat_template_tool_use = gr.Textbox( label = "Chat Prompt (tool_use)", interactive = False, lines = 20, show_copy_button = True, ) rendered_chat_template_rag = gr.Textbox( label = "Chat Prompt (rag)", interactive = False, lines = 20, show_copy_button = True, ) model_info = gr.State( value = {}, ) @gr.on( triggers = [ hf_search.submit, ], inputs = [ hf_search, ], outputs = [ hf_branch, ], show_api = False, ) def get_branches( repo: str, oauth_token: gr.OAuthToken | None = None, ): branches = [] try: refs = hfapi.list_repo_refs( repo, token = oauth_token.token if oauth_token else False, ) branches = [b.name for b in refs.branches] open_prs = hfapi.get_repo_discussions( repo, discussion_type = "pull_request", discussion_status = "open", token = oauth_token.token if oauth_token else False, ) branches += [pr.git_reference for pr in open_prs] except Exception as e: pass return { hf_branch: gr.Dropdown( branches or None, value = "main" if "main" in branches else None, ), } @gr.on( triggers = [ pr_title.input, ], inputs = [ pr_title, ], outputs = [ pr_submit, ], show_api = False, ) def enable_pr_submit( title: str, ): return gr.Button( interactive = bool(title) ) @gr.on( triggers = [ pr_preview_tab.select, ], inputs = [ model_info, pr_title, pr_description, chat_template, chat_template_tool_use, chat_template_rag, inverse_template, ], outputs = [ pr_preview_title, pr_preview_description, pr_preview_diff, ], show_api = False, ) def render_pr_preview( info: dict, title: str, description: str, template: str, template_tool_use: str, template_rag: str, template_inverse: str, ): changes = [] org_template = "" org_template_inverse = "" org_template_tool_use = "" org_template_rag = "" for config_file_name in (ModelFiles.CHAT_TEMPLATE_JSON, ModelFiles.TOKENIZER_CONFIG): config_file = info.get(config_file_name, {}) org_config = config_file.get("data") org_content = config_file.get("content") if org_content and ("chat_template" in org_content or not org_template): tokenizer_config = TokenizerConfig(org_content) org_template = tokenizer_config.chat_templates.get("default") or "" org_template_tool_use = tokenizer_config.chat_templates.get("tool_use") or "" org_template_rag = tokenizer_config.chat_templates.get("rag") or "" # org_template_inverse = tokenizer_config.inverse_template or "" tokenizer_config.chat_templates["default"] = template tokenizer_config.chat_templates["tool_use"] = template_tool_use tokenizer_config.chat_templates["rag"] = template_rag # tokenizer_config.inverse_template = template_inverse new_config = tokenizer_config.json(get_json_indent(org_config)) if org_config.endswith("\n"): new_config += "\n" changes += [ (token if token[1] in ("-", "+", "@") else token[1:].replace("\t", "\u21e5").replace("\r\n", "\u240d\u240a\r\n").replace("\r", "\u240d\r").replace("\n", "\u240a\n"), token[0] if token[0] != " " else None) # .replace(" ", "\u2423") for token in unified_diff(new_config.splitlines(keepends = True), org_config.splitlines(keepends = True), fromfile = config_file_name, tofile = config_file_name) ] tokenizer_chat_template = info.get(ModelFiles.TOKENIZER_CHAT_TEMPLATE, {}) org_template = tokenizer_chat_template.get("data", org_template) tokenizer_inverse_template = info.get(ModelFiles.TOKENIZER_INVERSE_TEMPLATE, {}) org_template_inverse = tokenizer_inverse_template.get("data", org_template_inverse) if org_template or template: changes += character_diff(f"Default Template{f' ({ModelFiles.TOKENIZER_CHAT_TEMPLATE})' if tokenizer_chat_template else ''}", org_template, template) if org_template_inverse or template_inverse: changes += character_diff(f"Inverse Template{f' ({ModelFiles.TOKENIZER_INVERSE_TEMPLATE})' if tokenizer_inverse_template else ''}", org_template_inverse, template_inverse) if org_template_tool_use or template_tool_use: changes += character_diff("Tool Use Template", org_template_tool_use, template_tool_use) if org_template_rag or template_rag: changes += character_diff("RAG Template", org_template_rag, template_rag) return title, description, changes @gr.on( triggers = [ pr_submit.click, ], inputs = [ hf_search, hf_branch, model_info, pr_title, pr_description, chat_template, chat_template_tool_use, chat_template_rag, inverse_template, ], outputs = [ model_info, hf_branch, pr_title, pr_preview_title, pr_description, pr_submit, pr_tabs, ], show_api = False, ) def submit_pull_request( repo: str, branch: str | None, info: dict, title: str, description: str, template: str, template_tool_use: str, template_rag: str, template_inverse: str, progress = gr.Progress(track_tqdm = True), oauth_token: gr.OAuthToken | None = None, ): operations = [] pr_branch = branch if branch.startswith("refs/pr/") else None chat_template_file = info.get(ModelFiles.CHAT_TEMPLATE_JSON, {}) if org_chat_template := chat_template_file.get("data"): tokenizer_config = TokenizerConfig(chat_template_file.get("content")) tokenizer_config.chat_templates["default"] = template tokenizer_config.chat_templates["tool_use"] = template_tool_use tokenizer_config.chat_templates["rag"] = template_rag # tokenizer_config.inverse_template = template_inverse new_chat_template = tokenizer_config.json(get_json_indent(org_chat_template)) if org_chat_template.endswith("\n"): new_chat_template += "\n" if org_chat_template != new_chat_template: operations.append(CommitOperationAdd(ModelFiles.CHAT_TEMPLATE_JSON, new_chat_template.encode("utf-8"))) tokenizer_file = info.get(ModelFiles.TOKENIZER_CONFIG, {}) if org_config := tokenizer_file.get("data"): tokenizer_content = tokenizer_file.get("content") tokenizer_config = TokenizerConfig(tokenizer_content) if "chat_template" in tokenizer_content or not chat_template_file: tokenizer_config.chat_templates["default"] = template tokenizer_config.chat_templates["tool_use"] = template_tool_use tokenizer_config.chat_templates["rag"] = template_rag # tokenizer_config.inverse_template = template_inverse new_config = tokenizer_config.json(get_json_indent(org_config)) if org_config.endswith("\n"): new_config += "\n" if org_config != new_config: operations.append(CommitOperationAdd(ModelFiles.TOKENIZER_CONFIG, new_config.encode("utf-8"))) tokenizer_chat_template = info.get(ModelFiles.TOKENIZER_CHAT_TEMPLATE, {}) if template_data := tokenizer_chat_template.get("data"): if template_data != template: operations.append(CommitOperationAdd(ModelFiles.TOKENIZER_CHAT_TEMPLATE, template.encode("utf-8"))) tokenizer_inverse_template = info.get(ModelFiles.TOKENIZER_INVERSE_TEMPLATE, {}) if template_data := tokenizer_inverse_template.get("data"): if template_data != template_inverse: operations.append(CommitOperationAdd(ModelFiles.TOKENIZER_INVERSE_TEMPLATE, template_inverse.encode("utf-8"))) if not operations: gr.Info("No changes to commit...") return gr.skip() try: commit = hfapi.create_commit( repo, operations, revision = branch, commit_message = title, commit_description = description, create_pr = False if pr_branch else True, parent_commit = info.get("parent_commit"), token = oauth_token.token if oauth_token else False, ) except Exception as e: gr.Warning( message = str(e), duration = None, title = "Error committing changes", ) return gr.skip() info["parent_commit"] = commit.oid if org_chat_template: chat_template_file["data"] = new_chat_template chat_template_file["content"] = json.loads(new_chat_template) if org_config: tokenizer_file["data"] = new_config tokenizer_file["content"] = json.loads(new_config) if tokenizer_chat_template: tokenizer_chat_template["data"] = template if tokenizer_inverse_template: tokenizer_inverse_template["data"] = template_inverse branches = [] try: refs = hfapi.list_repo_refs( repo, token = oauth_token.token if oauth_token else False, ) branches = [b.name for b in refs.branches] open_prs = hfapi.get_repo_discussions( repo, discussion_type = "pull_request", discussion_status = "open", token = oauth_token.token if oauth_token else False, ) branches += [pr.git_reference for pr in open_prs] except Exception as e: pass pr_created = commit.pr_revision if commit.pr_revision in branches else None gr.Info( message = "Successfully committed changes.", title = f"Pull Request {'Created' if pr_created else 'Updated'}", ) return { model_info: info, hf_branch: gr.skip() if pr_branch else gr.Dropdown( branches or None, value = pr_created or branch, ), pr_title: gr.skip() if pr_branch else gr.Textbox( value = None, placeholder = "Message" if pr_created else "Title", label = commit.commit_message if pr_created else None, show_label = True if pr_created else False, ), pr_preview_title: gr.skip() if pr_branch else gr.Textbox( label = commit.commit_message if pr_created else None, show_label = True if pr_created else False, ), pr_description: gr.Code( value = pr_description_default, ), pr_submit: gr.skip() if pr_branch else gr.Button( value = f"Commit to PR #{commit.pr_num}" if pr_created else "Create Pull Request", ), pr_tabs: gr.Tabs( selected = "edit", ), } @gr.on( triggers = [ hf_search.submit, hf_branch.change, ], outputs = [ pr_tabs, template_tabs, ], show_api = False, ) def switch_to_edit_tabs(): return gr.Tabs( selected = "edit", ), gr.Tabs( selected = "edit", ) @gr.on( triggers = [ chat_template.focus, chat_template_tool_use.focus, chat_template_rag.focus, inverse_template.focus, ], outputs = [ pr_tabs, ], show_api = False, ) def switch_to_edit_tab(): return gr.Tabs( selected = "edit", ) def template_data_from_model_info( repo: str, branch: str | None, oauth_token: gr.OAuthToken | None = None, ): try: info = hfapi.model_info( repo, revision = branch, expand = [ "config", "disabled", "gated", "gguf", "private", "widgetData", ], token = oauth_token.token if oauth_token else False, ) except Exception as e: gr.Warning( message = str(e), title = "Error loading model info", ) return {}, None, None, None, None, None, None templates = info.gguf.get("chat_template") if info.gguf else info.config.get("tokenizer_config", {}).get("chat_template") if info.config else None model_info = { "gguf": bool(info.gguf), "disabled": info.disabled, "gated": info.gated, "private": info.private, "chat_template": templates, } template_messages = example_values[0][1] template_tool_use = None template_rag = None template_inverse = None template_kwargs = { "add_generation_prompt": True, "clean_up_tokenization_spaces": False, "bos_token": "<|startoftext|>", "eos_token": "<|im_end|>", } if info.config: # template_inverse = info.config.get("tokenizer_config", {}).get("inverse_template") for k, v in info.config.get("tokenizer_config", {}).items(): if k != "chat_template": # and k != "inverse_template": template_kwargs[k] = v if info.widget_data: for data in info.widget_data: if "messages" in data: template_messages = json.dumps(data["messages"], ensure_ascii = False, indent = 2) break if isinstance(templates, list): templates = { template["name"]: template["template"] for template in templates } template_tool_use = templates.get("tool_use") template_rag = templates.get("rag") templates = templates.get("default") return model_info, json.dumps(template_kwargs, ensure_ascii = False, indent = 2), template_messages, templates, template_tool_use, template_rag, template_inverse def template_data_from_model_files( repo: str, branch: str | None, info: dict, progress = gr.Progress(track_tqdm = True), oauth_token: gr.OAuthToken | None = None, ): parent_commit = None write_access = False if info and oauth_token: if info.get("gguf"): gr.Warning("Repository contains GGUFs, use GGUF Editor if you want to commit changes...") elif info.get("disabled"): gr.Warning("Repository is disabled, committing changes is not possible...") elif (gated := info.get("gated")) or (private := info.get("private")): try: hfapi.auth_check( repo, token = oauth_token.token if oauth_token else False, ) except Exception as e: if gated: gr.Warning(f"Repository is gated with {gated} approval, you must request access to be able to make changes...") elif private: gr.Warning("Repository is private, you must use proper credentials to be able to make changes...") gr.Warning(str(e)) else: write_access = True else: write_access = True if write_access: if (write_access := hfapi.file_exists( repo, ModelFiles.TOKENIZER_CONFIG, revision = branch, token = oauth_token.token if oauth_token else False, )): try: commits = hfapi.list_repo_commits( repo, revision = branch, token = oauth_token.token if oauth_token else False, ) parent_commit = commits[0].commit_id if commits else None tokenizer_config_file = hfapi.hf_hub_download( repo, ModelFiles.TOKENIZER_CONFIG, revision = parent_commit or branch, token = oauth_token.token if oauth_token else False, ) tokenizer_chat_template = None if (hfapi.file_exists( repo, ModelFiles.TOKENIZER_CHAT_TEMPLATE, revision = branch, token = oauth_token.token if oauth_token else False, )): tokenizer_chat_template = hfapi.hf_hub_download( repo, ModelFiles.TOKENIZER_CHAT_TEMPLATE, revision = parent_commit or branch, token = oauth_token.token if oauth_token else False, ) tokenizer_inverse_template = None if (hfapi.file_exists( repo, ModelFiles.TOKENIZER_INVERSE_TEMPLATE, revision = branch, token = oauth_token.token if oauth_token else False, )): tokenizer_inverse_template = hfapi.hf_hub_download( repo, ModelFiles.TOKENIZER_INVERSE_TEMPLATE, revision = parent_commit or branch, token = oauth_token.token if oauth_token else False, ) except Exception as e: gr.Warning( message = str(e), title = "Error downloading template files", ) else: info["parent_commit"] = parent_commit if tokenizer_config_file: with open(tokenizer_config_file, "r", encoding = "utf-8") as fp: config_content = fp.read() info[ModelFiles.TOKENIZER_CONFIG] = { "data": config_content, "content": json.loads(config_content), } if tokenizer_chat_template: with open(tokenizer_chat_template, "r", encoding = "utf-8") as fp: template_data = fp.read() info[ModelFiles.TOKENIZER_CHAT_TEMPLATE] = { "data": template_data, } if tokenizer_inverse_template: with open(tokenizer_inverse_template, "r", encoding = "utf-8") as fp: template_data = fp.read() info[ModelFiles.TOKENIZER_INVERSE_TEMPLATE] = { "data": template_data, } else: gr.Warning(f"No {ModelFiles.TOKENIZER_CONFIG} found in repository...") if not info.get("chat_template"): try: chat_template_file = None if (hfapi.file_exists( repo, ModelFiles.CHAT_TEMPLATE_JSON, revision = branch, token = oauth_token.token if oauth_token else False, )): chat_template_file = hfapi.hf_hub_download( repo, ModelFiles.CHAT_TEMPLATE_JSON, revision = parent_commit or branch, token = oauth_token.token if oauth_token else False, ) except Exception as e: pass else: if chat_template_file: with open(chat_template_file, "r", encoding = "utf-8") as fp: template_data = fp.read() template_content = json.loads(template_data) info[ModelFiles.CHAT_TEMPLATE_JSON] = { "data": template_data, "content": template_content, } info["chat_template"] = template_content.get("chat_template") pr_details = None if branch and branch.startswith("refs/pr/"): pr_num = branch.split("/")[-1] if pr_num and pr_num.isdigit(): pr_details = hfapi.get_discussion_details( repo, int(pr_num), token = oauth_token.token if oauth_token else False, ) return { model_info: info, pr_group: gr.Accordion( visible = write_access, ), pr_title: gr.Textbox( value = None, placeholder = "Message" if pr_details else "Title", label = pr_details.title if pr_details else None, show_label = True if pr_details else False, ), pr_preview_title: gr.Textbox( label = pr_details.title if pr_details else None, show_label = True if pr_details else False, ), pr_description: gr.Code( value = pr_description_default, ), pr_submit: gr.Button( value = f"Commit to PR #{pr_details.num}" if pr_details else "Create Pull Request", ), chat_template: gr.skip() if ModelFiles.CHAT_TEMPLATE_JSON not in info else gr.Code( value = TokenizerConfig(info[ModelFiles.CHAT_TEMPLATE_JSON]["content"]).chat_templates.get("default"), ), # inverse_template: gr.skip() if ModelFiles.TOKENIZER_INVERSE_TEMPLATE not in info else gr.Code( # value = info[ModelFiles.TOKENIZER_INVERSE_TEMPLATE]["data"], # ), } def update_examples( settings: str, ): settings = json.loads(settings) examples = [] for example in example_values: x = example.copy() x0 = json.loads(x[0]) x0.update(settings) x[0] = json.dumps(x0, ensure_ascii = False, indent = 2) examples.append(x) return gr.Dataset( samples = examples, ) gr.on( fn = template_data_from_model_info, triggers = [ hf_search.submit, hf_branch.input, ], inputs = [ hf_search, hf_branch, ], outputs = [ model_info, chat_settings, chat_messages, chat_template, chat_template_tool_use, chat_template_rag, inverse_template, ], ).success( fn = update_examples, inputs = [ chat_settings, ], outputs = [ example_input.dataset, ], show_api = False, ).then( fn = template_data_from_model_files, inputs = [ hf_search, hf_branch, model_info, ], outputs = [ model_info, pr_group, pr_title, pr_preview_title, pr_description, pr_submit, chat_template, # inverse_template, ], show_api = False, ) @gr.on( triggers = [ render_tab.select, ], inputs = [ chat_settings, chat_messages, chat_template, chat_template_tool_use, chat_template_rag, inverse_template, ], outputs = [ rendered_chat_template, rendered_chat_template_tool_use, rendered_chat_template_rag, rendered_inverse_template, ], ) def render_chat_templates( settings: str, messages: str, template: str, template_tool_use: str | None = None, template_rag: str | None = None, template_inverse: str | None = None, ): try: settings = json.loads(settings) if settings else {} except Exception as e: gr.Warning( message = str(e), duration = None, title = "Template Settings Error", ) return gr.skip() try: messages = json.loads(messages) if messages else [] except Exception as e: gr.Warning( message = str(e), duration = None, title = "Template Messages Error", ) return gr.skip() if not isinstance(settings, dict): gr.Warning("Invalid Template Settings!") return gr.skip() if not messages or not isinstance(messages, list) or not isinstance(messages[0], dict) or "role" not in messages[0]: gr.Warning("No Template Messages!") return gr.skip() tools = settings.get("tools") documents = settings.get("documents") add_generation_prompt = settings.get("add_generation_prompt") cleanup_settings = [] for k in settings.keys(): if k.endswith("_side") or k.endswith("_token") or k.endswith("_tokens") or k == "clean_up_tokenization_spaces": continue cleanup_settings.append(k) for cleanup in cleanup_settings: del settings[cleanup] tokenizer = PreTrainedTokenizerBase(**settings) chat_output = None chat_tool_use_output = None chat_rag_output = None inverse_output = None try: chat_output = tokenizer.apply_chat_template(messages, tools = tools, documents = documents, chat_template = template, add_generation_prompt = add_generation_prompt, tokenize = False) except Exception as e: gr.Warning( message = str(e), duration = None, title = "Chat Template Error", ) try: chat_tool_use_output = tokenizer.apply_chat_template(messages, tools = tools or [], chat_template = template_tool_use, add_generation_prompt = add_generation_prompt, tokenize = False) if template_tool_use else None except Exception as e: gr.Warning( message = str(e), duration = None, title = "Tool Use Template Error", ) try: chat_rag_output = tokenizer.apply_chat_template(messages, documents = documents or [], chat_template = template_rag, add_generation_prompt = add_generation_prompt, tokenize = False) if template_rag else None except Exception as e: gr.Warning( message = str(e), duration = None, title = "RAG Template Error", ) try: inverse_output = tokenizer.apply_inverse_template(messages, inverse_template = template_inverse) if template_inverse else None except Exception as e: gr.Warning( message = str(e), duration = None, title = "Inverse Template Error", ) return chat_output, chat_tool_use_output, chat_rag_output, json.dumps(inverse_output, ensure_ascii = False, indent = 2) if inverse_output is not None else None if __name__ == "__main__": blocks.queue( max_size = 10, default_concurrency_limit = 10, ) blocks.launch()