Align tokenizer

special_tokens_map.json

@@ -1,6 +1,48 @@
 {
-  "bos_token": "<|startoftext|>",
-  "eos_token": "<|endoftext|>",
-  "pad_token": "<|pad|>",
-  "unk_token": "<|unk|>"
+  "additional_special_tokens": [
+    "<|eom|>",
+    "<|bom|>",
+    "<|system|>",
+    "<|user|>",
+    "<|assistant|>",
+    "<|tool|>",
+    "<documents>",
+    "</documents>",
+    "<tool_definitions>",
+    "</tool_definitions>",
+    "<active_output_modes>",
+    "</active_output_modes>",
+    "<citations>",
+    "</citations>",
+    "<tool_calls>",
+    "</tool_calls>"
+  ],
+  "bos_token": {
+    "content": "<|startoftext|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "<|pad|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "unk_token": {
+    "content": "<|unk|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
 }

tokenizer.json

@@ -38,6 +38,150 @@
       "rstrip": false,
       "normalized": false,
       "special": true
+    },
+    {
+      "id": 518,
+      "content": "<|eom|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    {
+      "id": 519,
+      "content": "<|bom|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    {
+      "id": 520,
+      "content": "<|system|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    {
+      "id": 521,
+      "content": "<|user|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    {
+      "id": 522,
+      "content": "<|assistant|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    {
+      "id": 523,
+      "content": "<|tool|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    {
+      "id": 524,
+      "content": "<documents>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    {
+      "id": 525,
+      "content": "</documents>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    {
+      "id": 526,
+      "content": "<tool_definitions>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    {
+      "id": 527,
+      "content": "</tool_definitions>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    {
+      "id": 528,
+      "content": "<active_output_modes>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    {
+      "id": 529,
+      "content": "</active_output_modes>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    {
+      "id": 530,
+      "content": "<citations>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    {
+      "id": 531,
+      "content": "</citations>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    {
+      "id": 532,
+      "content": "<tool_calls>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    {
+      "id": 533,
+      "content": "</tool_calls>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
     }
   ],
   "normalizer": {
@@ -133,6 +277,7 @@
     "end_of_word_suffix": null,
     "fuse_unk": true,
     "byte_fallback": true,
+    "ignore_merges": false,
     "vocab": {
       "<|pad|>": 0,
       "<|startoftext|>": 1,
@@ -652,22 +797,22 @@
       "<|mask_511|>": 515,
       "<|reserved_0|>": 516,
       "<|reserved_1|>": 517,
-      "<|reserved_2|>": 518,
-      "<|reserved_3|>": 519,
-      "<|reserved_4|>": 520,
-      "<|reserved_5|>": 521,
-      "<|reserved_6|>": 522,
-      "<|reserved_7|>": 523,
-      "<|reserved_8|>": 524,
-      "<|reserved_9|>": 525,
-      "<|reserved_10|>": 526,
-      "<|reserved_11|>": 527,
-      "<|reserved_12|>": 528,
-      "<|reserved_13|>": 529,
-      "<|reserved_14|>": 530,
-      "<|reserved_15|>": 531,
-      "<|reserved_16|>": 532,
-      "<|reserved_17|>": 533,
+      "<|eom|>": 518,
+      "<|bom|>": 519,
+      "<|system|>": 520,
+      "<|user|>": 521,
+      "<|assistant|>": 522,
+      "<|tool|>": 523,
+      "<documents>": 524,
+      "</documents>": 525,
+      "<tool_definitions>": 526,
+      "</tool_definitions>": 527,
+      "<active_output_modes>": 528,
+      "</active_output_modes>": 529,
+      "<citations>": 530,
+      "</citations>": 531,
+      "<tool_calls>": 532,
+      "</tool_calls>": 533,
       "<|reserved_18|>": 534,
       "<|reserved_19|>": 535,
       "<|reserved_20|>": 536,

tokenizer.model

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:02fd6530b8ede0eedd8e509fcab32da7b1dd04c8119f8498c787100f13112713
-size 1124742
+oid sha256:8b0df4fb43262c452ef37061951a06df4c63ca191d02a60ea08f14428af24376
+size 1124714

tokenizer_config.json

@@ -1,6 +1,7 @@
 {
   "add_bos_token": true,
   "add_eos_token": false,
+  "add_prefix_space": null,
   "added_tokens_decoder": {
     "0": {
       "content": "<|pad|>",
@@ -33,12 +34,159 @@
       "rstrip": false,
       "single_word": false,
       "special": true
+    },
+    "518": {
+      "content": "<|eom|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "519": {
+      "content": "<|bom|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "520": {
+      "content": "<|system|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "521": {
+      "content": "<|user|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "522": {
+      "content": "<|assistant|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "523": {
+      "content": "<|tool|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "524": {
+      "content": "<documents>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "525": {
+      "content": "</documents>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "526": {
+      "content": "<tool_definitions>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "527": {
+      "content": "</tool_definitions>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "528": {
+      "content": "<active_output_modes>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "529": {
+      "content": "</active_output_modes>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "530": {
+      "content": "<citations>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "531": {
+      "content": "</citations>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "532": {
+      "content": "<tool_calls>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "533": {
+      "content": "</tool_calls>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
     }
   },
+  "additional_special_tokens": [
+    "<|eom|>",
+    "<|bom|>",
+    "<|system|>",
+    "<|user|>",
+    "<|assistant|>",
+    "<|tool|>",
+    "<documents>",
+    "</documents>",
+    "<tool_definitions>",
+    "</tool_definitions>",
+    "<active_output_modes>",
+    "</active_output_modes>",
+    "<citations>",
+    "</citations>",
+    "<tool_calls>",
+    "</tool_calls>"
+  ],
   "bos_token": "<|startoftext|>",
-  "chat_template": "{% set bom_token='<|bom|>' %}{% set eom_token='<|eom|>' %}{% if messages[0]['role'] == 'system' %}{% set loop_messages = messages[1:] %}{% set system_message = messages[0]['content'] %}{% else %}{% set loop_messages = messages %}{% set system_message = 'You are a task following system, respond to the user\\'s input precisely. Do not add any additional information.' %}{% endif %}{{ bom_token + 'System: ' + system_message + eom_token + '' }}{% for message in loop_messages %}{% if (message['role'] == 'user') != (loop.index0 % 2 == 0) %}{{ raise_exception('Conversation roles must alternate user/assistant/user/assistant/...') }}{% endif %}{% set content = message['content'] %}{% if message['role'] == 'user' %}{{ bom_token + 'User: ' + content.strip() + eom_token }}{% elif message['role'] == 'system' %}{{ bom_token + 'System: ' + content.strip() + eom_token }}{% elif message['role'] == 'assistant' %}{{ bom_token + 'Assistant: ' + content.strip() + eom_token }}{% endif %}{% if message['role'] == 'assistant' and loop.index0 == loop.length - 1 %}{{ 'User: ' }}{% elif message['role'] == 'user' and loop.index0 == loop.length - 1 %}{{ '' + bom_token + 'Assistant:' }}{% else %}{{ '' }}{% endif %}{% endfor %}",
+  "chat_template": "{# Variables #}\n{% set ns = namespace(message_count=0, is_last_checked_defined=False) %}\n{##}\n{% set bom_str = bom_str or \"<|bom|>\" %}\n{% set eom_str = eom_str or \"<|eom|>\" %}\n{% set default_system_message = \"\" %}\n{##}\n{% set documents_prefix = \"<documents>\" %}\n{% set documents_suffix = \"</documents>\" %}\n{% set tool_definitions_prefix = \"<tool_definitions>\" %}\n{% set tool_definitions_suffix = \"</tool_definitions>\" %}\n{% set active_modes_prefix = \"<active_output_modes>\" %}\n{% set active_modes_suffix = \"</active_output_modes>\" %}\n{##}\n{% set tool_calls_prefix = \"<tool_calls>\" %}\n{% set tool_calls_suffix = \"</tool_calls>\" %}\n{% set citations_prefix = \"<citations>\" %}\n{% set citations_suffix = \"</citations>\" %}\n{##}\n{% if add_generation_prompt is not defined %}\n  {% set add_generation_prompt = True %}\n{% endif %}\n{% set role_to_predict = role_to_predict or \"assistant\" %}\n{% if messages|length > 0 and messages[0].role == \"system\" %}\n  {% set system_message = messages[0].content %}\n  {% set loop_messages = messages[1:] %}\n{% else %}\n  {% set system_message = default_system_message %}\n  {% set loop_messages = messages %}\n{% endif %}\n{##}\n{##}\n{# Macros #}\n{% macro handle_tool_definitions(tools) %}\n  {{- tool_definitions_prefix -}}\n  {{- \"\\n# Tools\" -}}\n  {{- \"\\n\\n## Functions\" -}}\n  {% for tool in tools %}\n    {% set _ = is_param_set(tool, field=\"type\") %}\n    {% set is_tool_type_set = ns.is_last_checked_defined %}\n    {% if is_tool_type_set %}\n      {% if tool.type == \"function\" %}\n        {% set tool = tool.function %}\n      {% else %}\n        {{ raise_exception(\"Currently, the only supported tool type is `function`\") }}\n      {% endif %}\n    {% endif %}\n    {{- \"\\n\\n\" + (tool|tojson(indent=2)) -}}\n  {% endfor %}\n  {{- \"\\n\" + tool_definitions_suffix -}}\n{% endmacro %}\n{##}\n{% macro handle_first_system_message(system_message, tools) %}\n  {{- bom_str + handle_role(\"system\") -}}\n  {% set _ = is_param_set(system_message) %}\n  {% set is_system_message_set = ns.is_last_checked_defined %}\n  {% if is_system_message_set %}\n    {{- system_message -}}\n  {% endif %}\n  {% set _ = is_param_set(tools, is_list=True) %}\n  {% set is_tools_set = ns.is_last_checked_defined %}\n  {% if is_tools_set %}\n    {% if system_message %}\n      {{- \"\\n\\n\" -}}\n    {% endif %}\n    {{- handle_tool_definitions(tools) -}}\n  {% endif %}\n  {% set ns.message_count = ns.message_count + 1 %}\n{% endmacro %}\n{##}\n{% macro handle_tool_calls(tool_calls) %}\n  {{- tool_calls_prefix + \"[\\n\" -}}\n  {% for tool_call in tool_calls %}\n    {% set _ = is_param_set(tool_call, field=\"function\") %}\n    {% set is_tool_call_function_set = ns.is_last_checked_defined %}\n    {% if is_tool_call_function_set %}\n      {%- set tool_call = tool_call.function %}\n    {%- endif %}\n    {% set arguments = tool_call.arguments %}\n    {% if arguments is not string %}\n      {%- set arguments = arguments|tojson -%}\n    {%- endif %}\n    {{ \"{\\\"name\\\": \\\"\" + tool_call.name + \"\\\", \\\"arguments\\\": \" + arguments + \"}\" -}}\n    {% if not loop.last %}\n      {{- \",\" }}\n    {% endif %}\n  {% endfor %}\n  {{- \"\\n]\" + tool_calls_suffix -}}\n{% endmacro %}\n{##}\n{% macro handle_documents(documents) %}\n  {{- documents_prefix -}}\n  {{- \"\\n# Documents\" -}}\n  {{- \"\\n\\nYou can use the following documents for reference:\" -}}\n  {% for doc in documents %}\n    {{- \"\\n\\n## Document ID: \" + loop.index0|string -}}\n    {% set _ = is_param_set(doc, field=\"title\") %}\n    {% set is_doc_title_set = ns.is_last_checked_defined %}\n    {% if is_doc_title_set %}\n      {{- \"\\nTitle: \" + doc.title -}}\n    {% endif %}\n    {% for key, value in doc.items() %}\n      {% if key not in [\"title\", \"text\"] %}\n        {{- \"\\n\" + key|title + \": \" + value|string -}}\n      {% endif %}\n    {% endfor %}\n    {{- \"\\nText: \" + doc.text -}}\n  {% endfor %}\n  {{- \"\\n\" + documents_suffix -}}\n{% endmacro %}\n{##}\n{% macro handle_knobs(knobs) %}\n  {{- active_modes_prefix -}}\n  {{- \"\\n# Active Modes\" -}}\n  {{ \"\\n\\nThe following modes configure the format or style of your responses. You should adhere to all currently\" -}}\n  {{ \" active modes simultaneously.\" -}}\n  {% if knobs.citation_mode == \"fast\" %}\n    {{- \"\\n\\n## Citation Mode\" -}}\n    {{- \"\\n\\nProvide a list of references only for the documents you base your response on. Format your response\" -}}\n    {{ \" with the original answer followed by a citation section. Use this template:\" -}}\n    {{ \" `{answer}\" + citations_prefix + \"DOCUMENT_IDS\" + citations_suffix + \"`, where DOCUMENT_IDS are the relevant document numbers\" -}}\n    {{ \" (e.g. [2, 5, 9]), or [] if the answer cannot be supported by the provided documents.\" -}}\n  {% endif %}\n  {% if knobs.response_format == \"json_object\" %}\n    {{- \"\\n\\n## JSON Mode\" -}}\n    {{ \"\\n\\nProvide your response in JSON format. Adhere strictly to any schema given by the user.\" -}}\n    {{ \" If an appropriate JSON format exists, use it without modification.\" -}}\n  {% endif %}\n  {{- \"\\n\" + active_modes_suffix -}}\n{% endmacro %}\n{##}\n{% macro get_last_user_index(messages) %}\n  {% set ns.last_user_index = 0 %}\n  {% for message in messages %}\n    {% if message.role == 'user' %}\n      {% set ns.last_user_index = loop.index0 %}\n    {% endif %}\n  {% endfor %}\n  {{- ns.last_user_index -}}\n{% endmacro %}\n{##}\n{% macro handle_last_system_message(documents, knobs, use_documents, use_knobs) %}\n  {{- bom_str + handle_role(\"system\") -}}\n  {% set macros_to_call = [] %}\n  {% set params_for_macros = [] %}\n  {% if use_documents %}\n    {% set macros_to_call = macros_to_call + [handle_documents] %}\n    {% set params_for_macros = params_for_macros + [[documents]] %}\n  {% endif %}\n  {% if use_knobs %}\n    {% set macros_to_call = macros_to_call + [handle_knobs] %}\n    {% set params_for_macros = params_for_macros + [[knobs]] %}\n  {% endif %}\n  {% for i in range(macros_to_call|length) %}\n    {% if i > 0 %}\n      {{- \"\\n\\n\" -}}\n    {% endif %}\n    {{- macros_to_call[i](*params_for_macros[i]) -}}\n  {% endfor %}\n  {% set ns.message_count = ns.message_count + 1 %}\n{% endmacro %}\n{##}\n{% macro handle_role(role, add_space=True) %}\n  {{- \"<|\" + role + \"|>\" -}}\n  {% if add_space %}\n    {{- \" \" -}}\n  {% endif %}\n{% endmacro %}\n{##}\n{% macro is_param_set(param, field=none, is_list=False) %}\n  {% if field is not none %}\n    {% if field in param %}\n      {% set param = param[field] %}\n    {% else %}\n      {% set param = none %}\n    {% endif %}\n  {% endif %}\n  {% set is_defined = param is defined and param is not none %}\n  {% if is_list %}\n    {% set ns.is_last_checked_defined = is_defined and param|length > 0 %}\n  {% else %}\n    {% set ns.is_last_checked_defined = is_defined %}\n  {% endif %}\n{% endmacro %}\n{##}\n{##}\n{# Template #}\n{{- \"<|startoftext|>\" -}}\n{% set _ = is_param_set(system_message) %}\n{% set is_system_message_set = ns.is_last_checked_defined %}\n{% set _ = is_param_set(tools, is_list=True) %}\n{% set is_tools_set = ns.is_last_checked_defined %}\n{% set has_system_message = (is_system_message_set or is_tools_set) %}\n{% if has_system_message %}\n  {{- handle_first_system_message(system_message, tools) -}}\n{% endif %}\n{% set last_user_index = get_last_user_index(loop_messages)|int %}\n{% for message in loop_messages %}\n  {% if loop.index0 == last_user_index %}\n    {% set _ = is_param_set(documents, is_list=True) %}\n    {% set use_documents = ns.is_last_checked_defined %}\n    {% set _ = is_param_set(knobs) %}\n    {% set use_knobs = ns.is_last_checked_defined and knobs.is_set %}\n    {% set add_last_system_message = use_documents or use_knobs %}\n    {% if add_last_system_message %}\n      {% if ns.message_count > 0 %}\n        {{- eom_str -}}\n      {% endif %}\n      {{- handle_last_system_message(documents, knobs, use_documents, use_knobs) -}}\n    {% endif %}\n  {% endif %}\n  {% set role = message.role %}\n  {% set _ = is_param_set(message, field=\"name\") %}\n  {% set is_message_name_set = ns.is_last_checked_defined %}\n  {% if is_message_name_set %}\n    {% set message_prefix = handle_role(role) + \"(\" + message.name + \")\" %}\n  {% else %}\n    {% set message_prefix = handle_role(role) %}\n  {% endif %}\n  {% set content = (message.content or \"\") %}\n  {% if content is not string %}\n    {% set content = content|tojson %}\n  {% endif %}\n  {% if ns.message_count > 0 %}\n    {{- eom_str -}}\n  {% endif %}\n  {{- bom_str + message_prefix + content -}}\n  {% set _ = is_param_set(message, field=\"tool_calls\", is_list=True) %}\n  {% set is_tool_calls_set = ns.is_last_checked_defined %}\n  {% if role == \"assistant\" and is_tool_calls_set %}\n    {{- handle_tool_calls(message.tool_calls) -}}\n  {% endif %}\n  {% set _ = is_param_set(message, field=\"citations\", is_list=True) %}\n  {% set is_citations_set = ns.is_last_checked_defined %}\n  {% if role == \"assistant\" and is_citations_set %}\n    {{- citations_prefix + message.citations|map(attribute=\"document_id\")|list|string + citations_suffix -}}\n  {% endif %}\n  {% set ns.message_count = ns.message_count + 1 %}\n{% endfor %}\n{% if add_generation_prompt %}\n  {% if ns.message_count > 0 %}\n    {{- eom_str -}}\n  {% endif %}\n  {{- bom_str + handle_role(role_to_predict, add_space=False) -}}\n  {% set _ = is_param_set(generation_preamble) %}\n  {% set is_generation_preamble_set = ns.is_last_checked_defined %}\n  {% if is_generation_preamble_set and generation_preamble.strip() != \"\" %}\n    {{- \" \" + generation_preamble -}}\n  {% endif %}\n  {% set ns.message_count = ns.message_count + 1 %}\n{% else %}\n  {% if ns.message_count > 0 %}\n    {{- eom_str -}}\n  {% endif %}\n{% endif %}\n",
   "clean_up_tokenization_spaces": false,
   "eos_token": "<|endoftext|>",
+  "legacy": true,
   "model_max_length": 1000000000000000019884624838656,
   "pad_token": "<|pad|>",
   "spaces_between_special_tokens": false,