From 7018c6fde1384f7115f46b9cbcc89bdcef0b4a2f Mon Sep 17 00:00:00 2001 From: Wing Lian Date: Fri, 15 Sep 2023 00:12:33 -0400 Subject: [PATCH 01/16] use fastchat conversations template --- src/axolotl/prompters.py | 72 +++++++++++----------------------------- 1 file changed, 20 insertions(+), 52 deletions(-) diff --git a/src/axolotl/prompters.py b/src/axolotl/prompters.py index d6d14c3694..7c66c8d8f7 100644 --- a/src/axolotl/prompters.py +++ b/src/axolotl/prompters.py @@ -1,9 +1,10 @@ """Module containing prompters""" -import dataclasses import logging from enum import Enum, auto -from typing import Generator, List, Optional, Tuple, Union +from typing import Generator, Optional, Union + +from fastchat.conversation import Conversation, get_conv_template LOG = logging.getLogger("axolotl") IGNORE_TOKEN_ID = -100 @@ -222,45 +223,6 @@ class SeparatorStyle(Enum): DOLLY = auto() -# TODO clean this 💩 up -@dataclasses.dataclass -class Conversation: - """A class that keeps all conversation history.""" - - system: str - roles: List[str] - messages: List[List[str]] - offset: int - sep_style: SeparatorStyle = SeparatorStyle.SINGLE - sep: str = "###" - sep2: Optional[str] = None - - def get_prompt(self) -> Generator[Tuple[str, str], None, None]: - # seps = [self.sep, self.sep2] - preamble = self.system + self.sep - yield ("SYSTEM:", preamble) - for _, (role, message) in enumerate(self.messages): - if message: - yield (role + ":", " " + message) - else: - LOG.warning(f"role with empty message: {role}") - yield (role + ":", "") - - def copy(self): - return Conversation( - system=self.system, - roles=self.roles, - messages=[[x, y] for x, y in self.messages], - offset=self.offset, - sep_style=self.sep_style, - sep=self.sep, - sep2=self.sep2, - ) - - def append_message(self, role, message): - self.messages.append([role, message]) - - SHAREGPT_ASSERTION_FAILED_ROLE = ( "Role did not alternate between turns (gpt and human). Please check your data." ) @@ -271,28 +233,34 @@ class ShareGPTPrompter: # pylint: disable=too-few-public-methods A prompter that generates prompts for the ShareGPT """ - def __init__(self, prompt_style=None, system_prompt: Optional[str] = None): + def __init__( + self, + prompt_style=None, + system_prompt: Optional[str] = None, + conversation: Optional[Union[str, Conversation]] = None, + ): if prompt_style != PromptStyle.CHAT.value: raise ValueError( f"unsupported prompt_style for ShareGPTPrompter({prompt_style})" ) system: str = ( system_prompt - if system_prompt + if system_prompt is not None else ( "A chat between a curious user and an artificial intelligence assistant. " "The assistant gives helpful, detailed, and polite answers to the user's questions." ) ) - self._conversation = Conversation( - system=system, - roles=["USER", "ASSISTANT"], - messages=[], - offset=0, - sep_style=SeparatorStyle.TWO, - sep=" ", - sep2=" ", - ) + if conversation: + if isinstance(conversation, Conversation): + self._conversation = conversation + else: + self._conversation = get_conv_template(conversation) + else: + self._conversation = get_conv_template("vicuna_v1.1") + + if system: + self._conversation.system_message = system def build_prompt(self, source) -> Generator[str, None, None]: if len(source) < 2: From 6cab7507eff3175cc6e4efdb139b4d22da7e52bc Mon Sep 17 00:00:00 2001 From: Wing Lian Date: Fri, 15 Sep 2023 00:18:38 -0400 Subject: [PATCH 02/16] require fastchat (fschat) pip install --- requirements.txt | 2 ++ 1 file changed, 2 insertions(+) diff --git a/requirements.txt b/requirements.txt index 33a2157d96..a3e87dacc9 100644 --- a/requirements.txt +++ b/requirements.txt @@ -31,3 +31,5 @@ scipy scikit-learn==1.2.2 pynvml art +wandb +fschat From 6449cdac5ead911f5769036b05306e70e72919d2 Mon Sep 17 00:00:00 2001 From: Wing Lian Date: Fri, 15 Sep 2023 01:15:19 -0400 Subject: [PATCH 03/16] handle roles dynamically from conversation --- src/axolotl/prompt_tokenizers.py | 8 ++++++-- src/axolotl/prompters.py | 10 +--------- 2 files changed, 7 insertions(+), 11 deletions(-) diff --git a/src/axolotl/prompt_tokenizers.py b/src/axolotl/prompt_tokenizers.py index a74da71da5..68b5b04bc4 100644 --- a/src/axolotl/prompt_tokenizers.py +++ b/src/axolotl/prompt_tokenizers.py @@ -6,6 +6,7 @@ import logging from typing import Dict, List, Tuple, Union +from fastchat.conversation import Conversation from transformers import BatchEncoding, PreTrainedTokenizer from axolotl.prompters import IGNORE_TOKEN_ID @@ -352,12 +353,15 @@ def tokenize_prompt(self, prompt): result, current_len = tokenize_prompt_default() user_token = self._get_user_token() assistant_token = self._get_assistant_token() + conversation: Conversation = ( + self.prompter._conversation # pylint: disable=protected-access + ) try: for _, part in enumerate( self.prompter.build_prompt(self.get_conversation_thread(prompt)) ): if isinstance(part, tuple): - if part[0] == "USER:": + if part[0] == conversation.roles[0]: turn = part[0] + part[1] if not user_token else part[1] # this is still the user query, we should if not part[1].strip(): @@ -371,7 +375,7 @@ def tokenize_prompt(self, prompt): res["input_ids"] = [user_token, *res["input_ids"]] # everything from this is masked out from the labels labels = [IGNORE_TOKEN_ID] * len(res["input_ids"]) - elif part[0] == "ASSISTANT:": + elif part[0] == conversation.roles[1]: # TODO label assistant token/tokens w/ IGNORE_TOKEN_ID turn = part[0] + part[1] if not assistant_token else part[1] # this should be the assistant response, should end with an eos token diff --git a/src/axolotl/prompters.py b/src/axolotl/prompters.py index 7c66c8d8f7..67eebb09c5 100644 --- a/src/axolotl/prompters.py +++ b/src/axolotl/prompters.py @@ -1,7 +1,7 @@ """Module containing prompters""" import logging -from enum import Enum, auto +from enum import Enum from typing import Generator, Optional, Union from fastchat.conversation import Conversation, get_conv_template @@ -215,14 +215,6 @@ def build_prompt( yield res -class SeparatorStyle(Enum): - """Different separator style.""" - - SINGLE = auto() - TWO = auto() - DOLLY = auto() - - SHAREGPT_ASSERTION_FAILED_ROLE = ( "Role did not alternate between turns (gpt and human). Please check your data." ) From 374d5671c1d91c51fcc35b7ed1fd3e7141c29790 Mon Sep 17 00:00:00 2001 From: Wing Lian Date: Fri, 15 Sep 2023 01:39:34 -0400 Subject: [PATCH 04/16] tweak fastchat conversation with a monkeypatch to get individual turns --- .../fastchat_conversation_turns.py | 26 +++++++++++++++++++ src/axolotl/prompt_tokenizers.py | 10 +++++-- src/axolotl/prompters.py | 2 +- 3 files changed, 35 insertions(+), 3 deletions(-) create mode 100644 src/axolotl/monkeypatch/fastchat_conversation_turns.py diff --git a/src/axolotl/monkeypatch/fastchat_conversation_turns.py b/src/axolotl/monkeypatch/fastchat_conversation_turns.py new file mode 100644 index 0000000000..00a6442540 --- /dev/null +++ b/src/axolotl/monkeypatch/fastchat_conversation_turns.py @@ -0,0 +1,26 @@ +""" +monkeypatch to add a get_turns method +""" + +import logging +from typing import Generator, Tuple + +LOG = logging.getLogger("axolotl.monkeypatch.fastchat_conversation_turns") + + +def get_turns(self) -> Generator[Tuple[str, str], None, None]: + # seps = [self.sep, self.sep2] + preamble = self.system_message + self.sep + yield ("SYSTEM:", preamble) + for _, (role, message) in enumerate(self.messages): + if message: + yield (role + ":", " " + message) + else: + LOG.warning(f"role with empty message: {role}") + yield (role + ":", "") + + +def add_get_turns_to_conversation(): + import fastchat.conversation + + fastchat.conversation.Conversation.get_turns = get_turns diff --git a/src/axolotl/prompt_tokenizers.py b/src/axolotl/prompt_tokenizers.py index 68b5b04bc4..342ff64838 100644 --- a/src/axolotl/prompt_tokenizers.py +++ b/src/axolotl/prompt_tokenizers.py @@ -9,6 +9,9 @@ from fastchat.conversation import Conversation from transformers import BatchEncoding, PreTrainedTokenizer +from axolotl.monkeypatch.fastchat_conversation_turns import ( + add_get_turns_to_conversation, +) from axolotl.prompters import IGNORE_TOKEN_ID LOG = logging.getLogger("axolotl") @@ -19,6 +22,8 @@ LLAMA_DEFAULT_BOS_TOKEN = "" # nosec LLAMA_DEFAULT_UNK_TOKEN = "" # nosec +add_get_turns_to_conversation() + class InvalidDataException(Exception): """ @@ -361,7 +366,7 @@ def tokenize_prompt(self, prompt): self.prompter.build_prompt(self.get_conversation_thread(prompt)) ): if isinstance(part, tuple): - if part[0] == conversation.roles[0]: + if part[0] == conversation.roles[0] + ":": turn = part[0] + part[1] if not user_token else part[1] # this is still the user query, we should if not part[1].strip(): @@ -375,7 +380,7 @@ def tokenize_prompt(self, prompt): res["input_ids"] = [user_token, *res["input_ids"]] # everything from this is masked out from the labels labels = [IGNORE_TOKEN_ID] * len(res["input_ids"]) - elif part[0] == conversation.roles[1]: + elif part[0] == conversation.roles[1] + ":": # TODO label assistant token/tokens w/ IGNORE_TOKEN_ID turn = part[0] + part[1] if not assistant_token else part[1] # this should be the assistant response, should end with an eos token @@ -403,6 +408,7 @@ def tokenize_prompt(self, prompt): labels = [IGNORE_TOKEN_ID] * len(res["input_ids"]) else: LOG.warning(f"unhandled role: {part[0]}") + continue # pylint: disable=duplicate-code result, current_len = parse_tokenized_to_result( diff --git a/src/axolotl/prompters.py b/src/axolotl/prompters.py index 67eebb09c5..e999a7eb46 100644 --- a/src/axolotl/prompters.py +++ b/src/axolotl/prompters.py @@ -289,5 +289,5 @@ def build_prompt(self, source) -> Generator[str, None, None]: assert role == conv.roles[j % 2], SHAREGPT_ASSERTION_FAILED_ROLE conv.append_message(role, sentence["value"]) - for part in conv.get_prompt(): + for part in conv.get_turns(): yield part From d411c70fa6c095bdc8811cd11d73431bce451387 Mon Sep 17 00:00:00 2001 From: Wing Lian Date: Sun, 24 Sep 2023 11:57:46 -0400 Subject: [PATCH 05/16] fix up so it works with multiple conversation styles, and don't strip the turns --- .../fastchat_conversation_turns.py | 166 +++++++++++++++++- src/axolotl/prompt_tokenizers.py | 14 +- src/axolotl/prompters.py | 2 + 3 files changed, 166 insertions(+), 16 deletions(-) diff --git a/src/axolotl/monkeypatch/fastchat_conversation_turns.py b/src/axolotl/monkeypatch/fastchat_conversation_turns.py index 00a6442540..20a3356a27 100644 --- a/src/axolotl/monkeypatch/fastchat_conversation_turns.py +++ b/src/axolotl/monkeypatch/fastchat_conversation_turns.py @@ -5,22 +5,170 @@ import logging from typing import Generator, Tuple +from fastchat.conversation import SeparatorStyle + LOG = logging.getLogger("axolotl.monkeypatch.fastchat_conversation_turns") -def get_turns(self) -> Generator[Tuple[str, str], None, None]: - # seps = [self.sep, self.sep2] - preamble = self.system_message + self.sep - yield ("SYSTEM:", preamble) - for _, (role, message) in enumerate(self.messages): - if message: - yield (role + ":", " " + message) +def get_prompt(self) -> str: + ret = "" + for role, msg in self.get_turns(): + ret += role + msg + return ret + + +def get_turns( # pylint: disable=too-many-return-statements + self, +) -> Generator[Tuple[str, str], None, None]: + """Get the prompt for generation.""" + system_prompt = self.system_template.format(system_message=self.system_message) + if self.sep_style == SeparatorStyle.ADD_COLON_SINGLE: + yield "", system_prompt + self.sep + for role, message in self.messages: + if message: + yield role + ": ", message + self.sep + else: + yield role + ":", "" + return + if self.sep_style == SeparatorStyle.ADD_COLON_TWO: + seps = [self.sep, self.sep2] + yield "", system_prompt + seps[0] + for i, (role, message) in enumerate(self.messages): + if message: + yield role + ": ", message + seps[i % 2] + else: + yield role + ":", "" + return + if self.sep_style == SeparatorStyle.ADD_COLON_SPACE_SINGLE: + yield "", system_prompt + self.sep + for role, message in self.messages: + if message: + yield role + ": ", message + self.sep + else: + yield role + ": ", "" # must be end with a space + return + if self.sep_style == SeparatorStyle.ADD_NEW_LINE_SINGLE: + yield "", "" if system_prompt == "" else system_prompt + self.sep + for role, message in self.messages: + if message: + yield role + "\n", message + self.sep + else: + yield role + "\n", "" + return + if self.sep_style == SeparatorStyle.NO_COLON_SINGLE: + yield "", system_prompt + for role, message in self.messages: + if message: + yield role, message + self.sep + else: + yield role, "" + return + if self.sep_style == SeparatorStyle.NO_COLON_TWO: + seps = [self.sep, self.sep2] + yield "", system_prompt + for i, (role, message) in enumerate(self.messages): + if message: + yield role, message + seps[i % 2] + else: + yield role, "" + return + if self.sep_style == SeparatorStyle.RWKV: + yield "", system_prompt + for i, (role, message) in enumerate(self.messages): + if message: + yield role + ": ", message.replace("\r\n", "\n").replace( + "\n\n", "\n" + ) + "\n\n" + else: + yield role + ":", "" + return + if self.sep_style == SeparatorStyle.LLAMA2: + seps = [self.sep, self.sep2] + if self.system_message: + yield "", system_prompt + self.messages[0] + " " else: - LOG.warning(f"role with empty message: {role}") - yield (role + ":", "") + yield "", "[INST] " + self.messages[0] + " " + for i, (role, message) in enumerate(self.messages[1:]): + if message: + yield role + " ", message + seps[i % 2] + else: + yield role, "" + return + if self.sep_style == SeparatorStyle.CHATGLM: + # source: https://huggingface.co/THUDM/chatglm-6b/blob/1d240ba371910e9282298d4592532d7f0f3e9f3e/modeling_chatglm.py#L1302-L1308 + # source2: https://huggingface.co/THUDM/chatglm2-6b/blob/e186c891cf64310ac66ef10a87e6635fa6c2a579/modeling_chatglm.py#L926 + round_add_n = 1 if self.name == "chatglm2" else 0 + if system_prompt: + yield "", system_prompt + self.sep + + for i, (role, message) in enumerate(self.messages): + if i % 2 == 0: + yield "", f"[Round {i//2 + round_add_n}]{self.sep}" + + if message: + yield f"{role}：", f"{message}{self.sep}" + else: + yield f"{role}：", "" + return + if self.sep_style == SeparatorStyle.CHATML: + yield "", "" if system_prompt == "" else system_prompt + self.sep + "\n" + for role, message in self.messages: + if message: + yield role + "\n", message + self.sep + "\n" + else: + yield role + "\n", "" + return + if self.sep_style == SeparatorStyle.CHATINTERN: + # source: https://huggingface.co/internlm/internlm-chat-7b-8k/blob/bd546fa984b4b0b86958f56bf37f94aa75ab8831/modeling_internlm.py#L771 + seps = [self.sep, self.sep2] + yield "", system_prompt + for i, (role, message) in enumerate(self.messages): + prefix = "" if i % 2 == 0 else "" + if message: + yield prefix + role + ":", message + seps[i % 2] + "\n" + else: + yield role + ":", "" + return + if self.sep_style == SeparatorStyle.DOLLY: + seps = [self.sep, self.sep2] + yield "", system_prompt + for i, (role, message) in enumerate(self.messages): + if message: + suffix = "\n\n" if i % 2 == 1 else "" + yield role + ":\n", message + seps[i % 2] + suffix + else: + yield role + ":\n", "" + return + if self.sep_style == SeparatorStyle.PHOENIX: + yield "", system_prompt + for role, message in self.messages: + if message: + yield role + ": ", "" + message + "" + else: + yield role + ": " + "", "" + return + if self.sep_style == SeparatorStyle.ROBIN: + yield "", system_prompt + self.sep + for role, message in self.messages: + if message: + yield role + ":\n", message + self.sep + else: + yield role + ":\n", "" + return + if self.sep_style == SeparatorStyle.FALCON_CHAT: + if self.system_message: + yield "", system_prompt + self.sep + for role, message in self.messages: + if message: + yield role + ": ", message + self.sep + else: + yield role + ":", "" + else: + raise ValueError(f"Invalid style: {self.sep_style}") def add_get_turns_to_conversation(): import fastchat.conversation fastchat.conversation.Conversation.get_turns = get_turns + fastchat.conversation.Conversation.get_prompt = get_prompt diff --git a/src/axolotl/prompt_tokenizers.py b/src/axolotl/prompt_tokenizers.py index 342ff64838..31bb73c2f7 100644 --- a/src/axolotl/prompt_tokenizers.py +++ b/src/axolotl/prompt_tokenizers.py @@ -366,13 +366,13 @@ def tokenize_prompt(self, prompt): self.prompter.build_prompt(self.get_conversation_thread(prompt)) ): if isinstance(part, tuple): - if part[0] == conversation.roles[0] + ":": + if conversation.roles[0] in part[0]: turn = part[0] + part[1] if not user_token else part[1] # this is still the user query, we should if not part[1].strip(): LOG.warning(f"user turn has empty text: {prompt}") res = self._tokenize( - turn.strip(), + turn, add_eos_token=False, strip_bos_token=True, ) @@ -380,14 +380,14 @@ def tokenize_prompt(self, prompt): res["input_ids"] = [user_token, *res["input_ids"]] # everything from this is masked out from the labels labels = [IGNORE_TOKEN_ID] * len(res["input_ids"]) - elif part[0] == conversation.roles[1] + ":": + elif conversation.roles[1] in part[0]: # TODO label assistant token/tokens w/ IGNORE_TOKEN_ID turn = part[0] + part[1] if not assistant_token else part[1] # this should be the assistant response, should end with an eos token if not part[1].strip(): LOG.warning(f"assistant turn has empty text: {prompt}") res = self._tokenize( - turn.strip(), + turn, add_eos_token=True, strip_bos_token=True, ) @@ -398,11 +398,11 @@ def tokenize_prompt(self, prompt): ] # not masked out from labels labels = copy.deepcopy(res["input_ids"]) - elif part[0] == "SYSTEM:": - part = part[1] # Ignore the system role from preamble + elif part[0] == "": + turn = part[1] # this is only ever the first part, should include the bos token and the user query res = self._tokenize( - part.strip(), add_eos_token=False, strip_bos_token=False + turn, add_eos_token=False, strip_bos_token=False ) # everything from this is masked out from the labels labels = [IGNORE_TOKEN_ID] * len(res["input_ids"]) diff --git a/src/axolotl/prompters.py b/src/axolotl/prompters.py index e999a7eb46..afb1bb4a1c 100644 --- a/src/axolotl/prompters.py +++ b/src/axolotl/prompters.py @@ -290,4 +290,6 @@ def build_prompt(self, source) -> Generator[str, None, None]: conv.append_message(role, sentence["value"]) for part in conv.get_turns(): + if part[0] and not part[1]: + LOG.warning(f"role with empty message: {part[0]}") yield part From 8d64467d8730c287d271eff2eb53db6bdb6c513e Mon Sep 17 00:00:00 2001 From: Wing Lian Date: Mon, 25 Sep 2023 12:48:27 -0400 Subject: [PATCH 06/16] fix sharegpt fixture now that we're using a more correct tokenization --- src/axolotl/utils/tokenization.py | 1 + tests/fixtures/conversation.tokenized.json | 2 +- 2 files changed, 2 insertions(+), 1 deletion(-) diff --git a/src/axolotl/utils/tokenization.py b/src/axolotl/utils/tokenization.py index 4746ceddef..2b7608b636 100644 --- a/src/axolotl/utils/tokenization.py +++ b/src/axolotl/utils/tokenization.py @@ -33,5 +33,6 @@ def check_example_labels(example, tokenizer, text_only=False): LOG.info(" ".join(colored_tokens)) LOG.info("\n\n\n") + print(" ".join(colored_tokens)) return " ".join(colored_tokens) diff --git a/tests/fixtures/conversation.tokenized.json b/tests/fixtures/conversation.tokenized.json index 0ac93713bc..ea4fca0990 100644 --- a/tests/fixtures/conversation.tokenized.json +++ b/tests/fixtures/conversation.tokenized.json @@ -1 +1 @@ -{"input_ids": [1, 319, 13563, 1546, 263, 12758, 1404, 322, 385, 23116, 21082, 20255, 29889, 450, 20255, 4076, 8444, 29892, 13173, 29892, 322, 1248, 568, 6089, 304, 278, 1404, 29915, 29879, 5155, 29889, 3148, 1001, 29901, 920, 1033, 474, 2334, 263, 29086, 705, 11356, 5687, 393, 3667, 4637, 21531, 20159, 304, 4505, 1045, 3163, 29973, 319, 1799, 9047, 13566, 29901, 739, 29915, 29879, 1950, 304, 671, 21531, 20159, 304, 4505, 341, 1475, 705, 11356, 29892, 541, 372, 723, 5517, 367, 3755, 4280, 29889, 838, 5869, 293, 20159, 338, 263, 5443, 310, 23964, 393, 11898, 278, 4426, 310, 8162, 393, 526, 21634, 1090, 9126, 316, 689, 800, 29892, 1316, 408, 16116, 292, 322, 289, 2548, 29889, 512, 278, 3030, 310, 341, 1475, 705, 11356, 29892, 445, 1033, 367, 1304, 304, 27599, 278, 3829, 310, 278, 3748, 7613, 322, 8161, 278, 14354, 310, 29086, 2729, 373, 3058, 25002, 15939, 1934, 29889, 2398, 29892, 16049, 445, 723, 1996, 263, 6483, 8004, 310, 1716, 21531, 20159, 322, 278, 3748, 310, 341, 1475, 705, 11356, 29889, 739, 29915, 29879, 451, 1554, 393, 508, 367, 10824, 297, 263, 2560, 1234, 29889, 2, 3148, 1001, 29901, 1033, 366, 2367, 592, 278, 330, 391, 310, 920, 372, 1033, 2466, 29973, 319, 1799, 9047, 13566, 29901, 18585, 29889, 450, 6996, 2969, 723, 367, 304, 671, 21531, 20159, 304, 27599, 278, 3829, 310, 278, 341, 1475, 705, 11356, 3748, 7613, 322, 8161, 278, 14354, 310, 29086, 2729, 373, 3058, 25002, 15939, 1934, 29889, 910, 1033, 25135, 15783, 278, 3748, 7613, 408, 263, 25002, 2913, 322, 773, 25002, 15939, 1934, 1316, 408, 3632, 327, 2270, 6471, 322, 350, 9890, 3694, 304, 10115, 2472, 1048, 278, 4423, 310, 29086, 373, 278, 7613, 29889, 2398, 29892, 408, 306, 5276, 1434, 29892, 16049, 445, 723, 1996, 263, 6483, 8004, 310, 1716, 21531, 20159, 322, 278, 3748, 310, 341, 1475, 705, 11356, 29892, 577, 372, 29915, 29879, 451, 1554, 393, 508, 367, 10824, 297, 263, 2560, 1234, 29889, 2, 3148, 1001, 29901, 2367, 592, 263, 2702, 1342, 319, 1799, 9047, 13566, 29901, 1094, 263, 2702, 1342, 29892, 7755, 591, 505, 263, 341, 1475, 705, 11356, 3748, 7613, 411, 278, 1494, 5285, 29901, 13, 28956, 13, 29871, 29896, 259, 29896, 259, 29896, 259, 29896, 259, 29896, 13, 29871, 29896, 29871, 1577, 29871, 1577, 29871, 1577, 259, 29896, 13, 29871, 29896, 29871, 1577, 29871, 1577, 29871, 1577, 259, 29896, 13, 29871, 29896, 29871, 1577, 29871, 1577, 29871, 1577, 259, 29896, 13, 29871, 29896, 259, 29896, 259, 29896, 259, 29896, 259, 29896, 13, 28956, 13, 797, 445, 1206, 29892, 278, 3748, 7613, 508, 367, 9875, 408, 263, 25002, 2913, 29892, 988, 1269, 6862, 373, 278, 7613, 16161, 304, 263, 1298, 297, 278, 2913, 29889, 450, 29086, 373, 278, 7613, 508, 367, 2714, 310, 408, 376, 5391, 267, 29908, 297, 278, 2913, 29892, 607, 526, 3291, 393, 526, 451, 760, 310, 278, 2913, 29889, 5293, 21531, 20159, 29892, 591, 508, 27599, 278, 3829, 310, 445, 2913, 322, 8161, 278, 14354, 310, 278, 29086, 2729, 373, 3058, 25002, 15939, 1934, 29889, 13, 13, 2831, 2777, 29892, 697, 982, 304, 437, 445, 338, 304, 10272, 278, 3632, 327, 2270, 6471, 310, 278, 2913, 29889, 15089, 327, 2270, 6471, 526, 263, 982, 310, 20766, 278, 26532, 297, 263, 2913, 29892, 322, 896, 508, 367, 1304, 304, 10115, 2472, 1048, 278, 14354, 310, 278, 29086, 373, 278, 341, 1475, 705, 11356, 7613, 29889, 512, 278, 1206, 310, 278, 7613, 2038, 29892, 591, 508, 10272, 278, 937, 3632, 327, 2270, 2318, 310, 278, 2913, 29892, 607, 16612, 278, 1353, 322, 376, 12181, 29908, 310, 278, 26532, 297, 278, 2913, 29889, 512, 445, 1206, 29892, 278, 937, 3632, 327, 2270, 2318, 723, 2649, 502, 393, 727, 526, 3023, 26532, 297, 278, 2913, 29892, 21050, 297, 263, 6862, 8267, 29889, 13, 13, 29933, 1463, 373, 445, 2472, 29892, 591, 508, 10115, 393, 278, 29086, 297, 278, 3748, 526, 5982, 297, 278, 3023, 26995, 310, 278, 7613, 29892, 408, 4318, 2400, 29901, 13, 28956, 13, 29871, 29896, 259, 29896, 259, 29896, 259, 29896, 259, 29896, 13, 29871, 29896, 29871, 341, 29871, 1577, 29871, 1577, 259, 29896, 13, 29871, 29896, 29871, 1577, 29871, 1577, 29871, 1577, 259, 29896, 13, 29871, 29896, 29871, 1577, 29871, 1577, 29871, 341, 259, 29896, 13, 29871, 29896, 259, 29896, 259, 29896, 259, 29896, 259, 29896, 13, 28956, 13, 2776, 3236, 29892, 445, 338, 925, 697, 1342, 310, 920, 21531, 20159, 1033, 367, 1304, 304, 4505, 341, 1475, 705, 11356, 29889, 1670, 526, 1784, 916, 1950, 13501, 29892, 322, 278, 2702, 4902, 310, 920, 304, 2334, 963, 723, 8839, 373, 278, 2702, 3748, 7613, 322, 278, 25002, 15939, 1934, 1641, 1304, 29889, 1094, 306, 5276, 1434, 29892, 16049, 263, 1650, 763, 445, 723, 1996, 263, 6483, 8004, 310, 1716, 21531, 20159, 322, 278, 3748, 310, 341, 1475, 705, 11356, 29889, 2, 3148, 1001, 29901, 10241, 474, 471, 4856, 411, 263, 6483, 8004, 310, 1716, 29892, 1033, 366, 5649, 278, 1021, 6964, 304, 592, 411, 393, 11833, 29973, 319, 1799, 9047, 13566, 29901, 18585, 29889, 17090, 366, 505, 263, 6483, 8004, 310, 1716, 21531, 20159, 322, 278, 3748, 310, 341, 1475, 705, 11356, 29892, 1244, 338, 920, 366, 1795, 671, 21531, 20159, 304, 4505, 263, 341, 1475, 705, 11356, 3748, 7613, 29901, 13, 13, 29896, 29889, 16314, 278, 341, 1475, 705, 11356, 3748, 7613, 408, 263, 25002, 2913, 29889, 7806, 6862, 373, 278, 7613, 16161, 304, 263, 1298, 297, 278, 2913, 29892, 322, 278, 29086, 373, 278, 7613, 508, 367, 2714, 310, 408, 376, 5391, 267, 29908, 297, 278, 2913, 29892, 607, 526, 3291, 393, 526, 451, 760, 310, 278, 2913, 29889, 13, 29906, 29889, 4803, 25002, 15939, 1934, 304, 10115, 2472, 1048, 278, 14354, 310, 278, 29086, 373, 278, 7613, 29889, 1670, 526, 1784, 1950, 25002, 15939, 1934, 393, 1033, 367, 1304, 363, 445, 29892, 1316, 408, 3632, 327, 2270, 6471, 29892, 350, 9890, 3694, 29892, 470, 28152, 3632, 3002, 29889, 13, 29941, 29889, 4803, 278, 2472, 10115, 1127, 515, 278, 25002, 15939, 1934, 304, 8161, 278, 14354, 310, 278, 29086, 373, 278, 7613, 29889, 1152, 1342, 29892, 565, 278, 937, 3632, 327, 2270, 2318, 310, 278, 2913, 14088, 393, 727, 526, 3023, 26532, 297, 278, 2913, 29892, 21050, 297, 263, 6862, 8267, 29892, 769, 591, 508, 10115, 393, 278, 29086, 526, 5982, 297, 278, 3023, 26995, 310, 278, 7613, 29889, 13, 29946, 29889, 4803, 445, 2472, 304, 4505, 278, 341, 1475, 705, 11356, 3748, 29889, 9038, 366, 1073, 278, 14354, 310, 278, 29086, 29892, 366, 508, 671, 445, 2472, 304, 4772, 963, 322, 2821, 278, 7613, 29889, 13, 13, 2776, 3236, 29892, 445, 338, 925, 697, 1950, 2948, 304, 773, 21531, 20159, 304, 4505, 341, 1475, 705, 11356, 29889, 1670, 1122, 367, 916, 5837, 304, 437, 372, 29892, 8679, 373, 278, 2702, 3748, 7613, 322, 278, 25002, 15939, 1934, 1641, 1304, 29889, 2398, 29892, 445, 881, 2367, 366, 263, 2498, 2969, 310, 920, 21531, 20159, 508, 367, 1304, 304, 4505, 341, 1475, 705, 11356, 29889, 2, 3148, 1001, 29901, 920, 723, 474, 437, 372, 411, 28152, 3632, 3002, 470, 738, 310, 278, 916, 25002, 15939, 1934, 29973, 319, 1799, 9047, 13566, 29901, 1763, 671, 28152, 3632, 3002, 470, 738, 310, 278, 916, 25002, 15939, 1934, 304, 4505, 263, 341, 1475, 705, 11356, 3748, 7613, 29892, 366, 723, 1101, 278, 1021, 6996, 6576, 408, 5439, 2038, 29892, 541, 411, 777, 12651, 297, 278, 2702, 4902, 29889, 13, 13, 2831, 2777, 29892, 1244, 338, 920, 366, 1795, 671, 28152, 3632, 3002, 304, 4505, 263, 341, 1475, 705, 11356, 3748, 7613, 29901, 13, 13, 29896, 29889, 16314, 278, 341, 1475, 705, 11356, 3748, 7613, 408, 263, 25002, 2913, 29889, 7806, 6862, 373, 278, 7613, 16161, 304, 263, 1298, 297, 278, 2913, 29892, 322, 278, 29086, 373, 278, 7613, 508, 367, 2714, 310, 408, 376, 5391, 267, 29908, 297, 278, 2913, 29892, 607, 526, 3291, 393, 526, 451, 760, 310, 278, 2913, 29889, 13, 29906, 29889, 11796, 29872, 278, 28152, 3632, 3002, 310, 278, 2913, 29889, 9034, 9696, 3632, 3002, 338, 263, 982, 310, 20766, 278, 26532, 297, 263, 2913, 975, 931, 29892, 322, 372, 508, 367, 1304, 304, 10115, 2472, 1048, 278, 14354, 310, 278, 29086, 373, 278, 341, 1475, 705, 11356, 7613, 29889, 13, 29941, 29889, 4803, 278, 2472, 515, 278, 28152, 3632, 3002, 304, 8161, 278, 14354, 310, 278, 29086, 373, 278, 7613, 29889, 1152, 1342, 29892, 565, 278, 28152, 3632, 3002, 14088, 393, 727, 526, 3023, 26532, 297, 278, 2913, 29892, 21050, 297, 263, 6862, 8267, 29892, 769, 591, 508, 10115, 393, 278, 29086, 526, 5982, 297, 278, 3023, 26995, 310, 278, 7613, 29889, 13, 29946, 29889, 4803, 445, 2472, 304, 4505, 278, 341, 1475, 705, 11356, 3748, 29889, 9038, 366, 1073, 278, 14354, 310, 278, 29086, 29892, 366, 508, 671, 445, 2472, 304, 4772, 963, 322, 2821, 278, 7613, 29889, 13, 13, 2776, 3236, 29892, 445, 338, 925, 697, 1950, 2948, 304, 773, 28152, 3632, 3002, 304, 4505, 341, 1475, 705, 11356, 29889, 1670, 1122, 367, 916, 5837, 304, 437, 372, 29892, 8679, 373, 278, 2702, 3748, 7613, 322, 278, 2702, 4902, 310, 278, 28152, 3632, 3002, 16287, 29889, 2398, 29892, 445, 881, 2367, 366, 263, 2498, 2969, 310, 920, 28152, 3632, 3002, 508, 367, 1304, 304, 4505, 341, 1475, 705, 11356, 29889, 2, 3148, 1001, 29901, 920, 1033, 474, 2334, 445, 297, 3017, 29973, 319, 1799, 9047, 13566, 29901, 1763, 2334, 263, 341, 1475, 705, 11356, 899, 369, 393, 3913, 21531, 20159, 297, 5132, 29892, 366, 723, 817, 304, 437, 278, 1494, 29901, 13, 13, 29896, 29889, 16052, 263, 5132, 3577, 363, 21531, 20159, 29889, 1670, 526, 3196, 3987, 3625, 29892, 1316, 408, 402, 566, 2918, 29892, 360, 291, 952, 375, 29892, 470, 4560, 7354, 29899, 29873, 1388, 29889, 4525, 9741, 3867, 14009, 322, 848, 12286, 363, 20602, 25002, 15939, 1934, 29892, 1316, 408, 3632, 327, 2270, 6471, 29892, 350, 9890, 3694, 29892, 470, 28152, 3632, 3002, 29889, 13, 29906, 29889, 16314, 278, 341, 1475, 705, 11356, 3748, 7613, 408, 263, 25002, 2913, 29889, 910, 1033, 367, 2309, 773, 263, 848, 3829, 4944, 491, 278, 21531, 20159, 3577, 29892, 1316, 408, 263, 3053, 506, 616, 4280, 470, 263, 478, 2035, 29367, 29899, 29934, 4512, 4280, 29889, 7806, 6862, 373, 278, 7613, 723, 3928, 304, 263, 1298, 297, 278, 2913, 29892, 322, 278, 29086, 373, 278, 7613, 723, 367, 9875, 408, 376, 5391, 267, 29908, 297, 278, 2913, 29889, 13, 29941, 29889, 11796, 29872, 278, 25002, 15939, 1934, 310, 278, 2913, 29889, 910, 1033, 367, 2309, 773, 14009, 4944, 491, 278, 21531, 20159, 3577, 29889, 1152, 1342, 29892, 366, 1033, 10272, 278, 28152, 3632, 3002, 310, 278, 2913, 773, 278, 28152, 3632, 3002, 5687, 29889, 13, 29946, 29889, 4803, 278, 2472, 515, 278, 25002, 15939, 1934, 304, 8161, 278, 14354, 310, 278, 29086, 373, 278, 7613, 29889, 910, 1033, 367, 2309, 773, 263, 10296, 310, 19475, 24481, 322, 8720, 29889, 1152, 1342, 29892, 565, 278, 28152, 3632, 3002, 14088, 393, 727, 526, 3023, 26532, 297, 278, 2913, 29892, 21050, 297, 263, 6862, 8267, 29892, 769, 366, 1033, 671, 263, 8424, 310, 775, 304, 10115, 393, 278, 29086, 526, 5982, 297, 278, 3023, 26995, 310, 278, 7613, 29889, 13, 29945, 29889, 4803, 445, 2472, 304, 4505, 278, 341, 1475, 705, 11356, 3748, 29889, 9038, 366, 1073, 278, 14354, 310, 278, 29086, 29892, 366, 1033, 671, 263, 8424, 310, 775, 304, 6336, 2828, 373, 278, 9109, 25256, 322, 2821, 278, 7613, 29889, 13, 13, 2776, 3236, 29892, 445, 338, 925, 697, 1950, 2948, 304, 16049, 263, 341, 1475, 705, 11356, 899, 369, 393, 3913, 21531, 20159, 297, 5132, 29889, 1670, 1122, 367, 916, 5837, 304, 437, 372, 29892, 8679, 373, 278, 2702, 4902, 310, 278, 5314, 29889, 2398, 29892, 445, 881, 2367, 366, 263, 2498, 2969, 310, 920, 304, 679, 4687, 411, 1316, 263, 2060, 29889, 2], "attention_mask": [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], "labels": [-100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, 319, 1799, 9047, 13566, 29901, 739, 29915, 29879, 1950, 304, 671, 21531, 20159, 304, 4505, 341, 1475, 705, 11356, 29892, 541, 372, 723, 5517, 367, 3755, 4280, 29889, 838, 5869, 293, 20159, 338, 263, 5443, 310, 23964, 393, 11898, 278, 4426, 310, 8162, 393, 526, 21634, 1090, 9126, 316, 689, 800, 29892, 1316, 408, 16116, 292, 322, 289, 2548, 29889, 512, 278, 3030, 310, 341, 1475, 705, 11356, 29892, 445, 1033, 367, 1304, 304, 27599, 278, 3829, 310, 278, 3748, 7613, 322, 8161, 278, 14354, 310, 29086, 2729, 373, 3058, 25002, 15939, 1934, 29889, 2398, 29892, 16049, 445, 723, 1996, 263, 6483, 8004, 310, 1716, 21531, 20159, 322, 278, 3748, 310, 341, 1475, 705, 11356, 29889, 739, 29915, 29879, 451, 1554, 393, 508, 367, 10824, 297, 263, 2560, 1234, 29889, 2, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, 319, 1799, 9047, 13566, 29901, 18585, 29889, 450, 6996, 2969, 723, 367, 304, 671, 21531, 20159, 304, 27599, 278, 3829, 310, 278, 341, 1475, 705, 11356, 3748, 7613, 322, 8161, 278, 14354, 310, 29086, 2729, 373, 3058, 25002, 15939, 1934, 29889, 910, 1033, 25135, 15783, 278, 3748, 7613, 408, 263, 25002, 2913, 322, 773, 25002, 15939, 1934, 1316, 408, 3632, 327, 2270, 6471, 322, 350, 9890, 3694, 304, 10115, 2472, 1048, 278, 4423, 310, 29086, 373, 278, 7613, 29889, 2398, 29892, 408, 306, 5276, 1434, 29892, 16049, 445, 723, 1996, 263, 6483, 8004, 310, 1716, 21531, 20159, 322, 278, 3748, 310, 341, 1475, 705, 11356, 29892, 577, 372, 29915, 29879, 451, 1554, 393, 508, 367, 10824, 297, 263, 2560, 1234, 29889, 2, -100, -100, -100, -100, -100, -100, -100, -100, 319, 1799, 9047, 13566, 29901, 1094, 263, 2702, 1342, 29892, 7755, 591, 505, 263, 341, 1475, 705, 11356, 3748, 7613, 411, 278, 1494, 5285, 29901, 13, 28956, 13, 29871, 29896, 259, 29896, 259, 29896, 259, 29896, 259, 29896, 13, 29871, 29896, 29871, 1577, 29871, 1577, 29871, 1577, 259, 29896, 13, 29871, 29896, 29871, 1577, 29871, 1577, 29871, 1577, 259, 29896, 13, 29871, 29896, 29871, 1577, 29871, 1577, 29871, 1577, 259, 29896, 13, 29871, 29896, 259, 29896, 259, 29896, 259, 29896, 259, 29896, 13, 28956, 13, 797, 445, 1206, 29892, 278, 3748, 7613, 508, 367, 9875, 408, 263, 25002, 2913, 29892, 988, 1269, 6862, 373, 278, 7613, 16161, 304, 263, 1298, 297, 278, 2913, 29889, 450, 29086, 373, 278, 7613, 508, 367, 2714, 310, 408, 376, 5391, 267, 29908, 297, 278, 2913, 29892, 607, 526, 3291, 393, 526, 451, 760, 310, 278, 2913, 29889, 5293, 21531, 20159, 29892, 591, 508, 27599, 278, 3829, 310, 445, 2913, 322, 8161, 278, 14354, 310, 278, 29086, 2729, 373, 3058, 25002, 15939, 1934, 29889, 13, 13, 2831, 2777, 29892, 697, 982, 304, 437, 445, 338, 304, 10272, 278, 3632, 327, 2270, 6471, 310, 278, 2913, 29889, 15089, 327, 2270, 6471, 526, 263, 982, 310, 20766, 278, 26532, 297, 263, 2913, 29892, 322, 896, 508, 367, 1304, 304, 10115, 2472, 1048, 278, 14354, 310, 278, 29086, 373, 278, 341, 1475, 705, 11356, 7613, 29889, 512, 278, 1206, 310, 278, 7613, 2038, 29892, 591, 508, 10272, 278, 937, 3632, 327, 2270, 2318, 310, 278, 2913, 29892, 607, 16612, 278, 1353, 322, 376, 12181, 29908, 310, 278, 26532, 297, 278, 2913, 29889, 512, 445, 1206, 29892, 278, 937, 3632, 327, 2270, 2318, 723, 2649, 502, 393, 727, 526, 3023, 26532, 297, 278, 2913, 29892, 21050, 297, 263, 6862, 8267, 29889, 13, 13, 29933, 1463, 373, 445, 2472, 29892, 591, 508, 10115, 393, 278, 29086, 297, 278, 3748, 526, 5982, 297, 278, 3023, 26995, 310, 278, 7613, 29892, 408, 4318, 2400, 29901, 13, 28956, 13, 29871, 29896, 259, 29896, 259, 29896, 259, 29896, 259, 29896, 13, 29871, 29896, 29871, 341, 29871, 1577, 29871, 1577, 259, 29896, 13, 29871, 29896, 29871, 1577, 29871, 1577, 29871, 1577, 259, 29896, 13, 29871, 29896, 29871, 1577, 29871, 1577, 29871, 341, 259, 29896, 13, 29871, 29896, 259, 29896, 259, 29896, 259, 29896, 259, 29896, 13, 28956, 13, 2776, 3236, 29892, 445, 338, 925, 697, 1342, 310, 920, 21531, 20159, 1033, 367, 1304, 304, 4505, 341, 1475, 705, 11356, 29889, 1670, 526, 1784, 916, 1950, 13501, 29892, 322, 278, 2702, 4902, 310, 920, 304, 2334, 963, 723, 8839, 373, 278, 2702, 3748, 7613, 322, 278, 25002, 15939, 1934, 1641, 1304, 29889, 1094, 306, 5276, 1434, 29892, 16049, 263, 1650, 763, 445, 723, 1996, 263, 6483, 8004, 310, 1716, 21531, 20159, 322, 278, 3748, 310, 341, 1475, 705, 11356, 29889, 2, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, 319, 1799, 9047, 13566, 29901, 18585, 29889, 17090, 366, 505, 263, 6483, 8004, 310, 1716, 21531, 20159, 322, 278, 3748, 310, 341, 1475, 705, 11356, 29892, 1244, 338, 920, 366, 1795, 671, 21531, 20159, 304, 4505, 263, 341, 1475, 705, 11356, 3748, 7613, 29901, 13, 13, 29896, 29889, 16314, 278, 341, 1475, 705, 11356, 3748, 7613, 408, 263, 25002, 2913, 29889, 7806, 6862, 373, 278, 7613, 16161, 304, 263, 1298, 297, 278, 2913, 29892, 322, 278, 29086, 373, 278, 7613, 508, 367, 2714, 310, 408, 376, 5391, 267, 29908, 297, 278, 2913, 29892, 607, 526, 3291, 393, 526, 451, 760, 310, 278, 2913, 29889, 13, 29906, 29889, 4803, 25002, 15939, 1934, 304, 10115, 2472, 1048, 278, 14354, 310, 278, 29086, 373, 278, 7613, 29889, 1670, 526, 1784, 1950, 25002, 15939, 1934, 393, 1033, 367, 1304, 363, 445, 29892, 1316, 408, 3632, 327, 2270, 6471, 29892, 350, 9890, 3694, 29892, 470, 28152, 3632, 3002, 29889, 13, 29941, 29889, 4803, 278, 2472, 10115, 1127, 515, 278, 25002, 15939, 1934, 304, 8161, 278, 14354, 310, 278, 29086, 373, 278, 7613, 29889, 1152, 1342, 29892, 565, 278, 937, 3632, 327, 2270, 2318, 310, 278, 2913, 14088, 393, 727, 526, 3023, 26532, 297, 278, 2913, 29892, 21050, 297, 263, 6862, 8267, 29892, 769, 591, 508, 10115, 393, 278, 29086, 526, 5982, 297, 278, 3023, 26995, 310, 278, 7613, 29889, 13, 29946, 29889, 4803, 445, 2472, 304, 4505, 278, 341, 1475, 705, 11356, 3748, 29889, 9038, 366, 1073, 278, 14354, 310, 278, 29086, 29892, 366, 508, 671, 445, 2472, 304, 4772, 963, 322, 2821, 278, 7613, 29889, 13, 13, 2776, 3236, 29892, 445, 338, 925, 697, 1950, 2948, 304, 773, 21531, 20159, 304, 4505, 341, 1475, 705, 11356, 29889, 1670, 1122, 367, 916, 5837, 304, 437, 372, 29892, 8679, 373, 278, 2702, 3748, 7613, 322, 278, 25002, 15939, 1934, 1641, 1304, 29889, 2398, 29892, 445, 881, 2367, 366, 263, 2498, 2969, 310, 920, 21531, 20159, 508, 367, 1304, 304, 4505, 341, 1475, 705, 11356, 29889, 2, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, 319, 1799, 9047, 13566, 29901, 1763, 671, 28152, 3632, 3002, 470, 738, 310, 278, 916, 25002, 15939, 1934, 304, 4505, 263, 341, 1475, 705, 11356, 3748, 7613, 29892, 366, 723, 1101, 278, 1021, 6996, 6576, 408, 5439, 2038, 29892, 541, 411, 777, 12651, 297, 278, 2702, 4902, 29889, 13, 13, 2831, 2777, 29892, 1244, 338, 920, 366, 1795, 671, 28152, 3632, 3002, 304, 4505, 263, 341, 1475, 705, 11356, 3748, 7613, 29901, 13, 13, 29896, 29889, 16314, 278, 341, 1475, 705, 11356, 3748, 7613, 408, 263, 25002, 2913, 29889, 7806, 6862, 373, 278, 7613, 16161, 304, 263, 1298, 297, 278, 2913, 29892, 322, 278, 29086, 373, 278, 7613, 508, 367, 2714, 310, 408, 376, 5391, 267, 29908, 297, 278, 2913, 29892, 607, 526, 3291, 393, 526, 451, 760, 310, 278, 2913, 29889, 13, 29906, 29889, 11796, 29872, 278, 28152, 3632, 3002, 310, 278, 2913, 29889, 9034, 9696, 3632, 3002, 338, 263, 982, 310, 20766, 278, 26532, 297, 263, 2913, 975, 931, 29892, 322, 372, 508, 367, 1304, 304, 10115, 2472, 1048, 278, 14354, 310, 278, 29086, 373, 278, 341, 1475, 705, 11356, 7613, 29889, 13, 29941, 29889, 4803, 278, 2472, 515, 278, 28152, 3632, 3002, 304, 8161, 278, 14354, 310, 278, 29086, 373, 278, 7613, 29889, 1152, 1342, 29892, 565, 278, 28152, 3632, 3002, 14088, 393, 727, 526, 3023, 26532, 297, 278, 2913, 29892, 21050, 297, 263, 6862, 8267, 29892, 769, 591, 508, 10115, 393, 278, 29086, 526, 5982, 297, 278, 3023, 26995, 310, 278, 7613, 29889, 13, 29946, 29889, 4803, 445, 2472, 304, 4505, 278, 341, 1475, 705, 11356, 3748, 29889, 9038, 366, 1073, 278, 14354, 310, 278, 29086, 29892, 366, 508, 671, 445, 2472, 304, 4772, 963, 322, 2821, 278, 7613, 29889, 13, 13, 2776, 3236, 29892, 445, 338, 925, 697, 1950, 2948, 304, 773, 28152, 3632, 3002, 304, 4505, 341, 1475, 705, 11356, 29889, 1670, 1122, 367, 916, 5837, 304, 437, 372, 29892, 8679, 373, 278, 2702, 3748, 7613, 322, 278, 2702, 4902, 310, 278, 28152, 3632, 3002, 16287, 29889, 2398, 29892, 445, 881, 2367, 366, 263, 2498, 2969, 310, 920, 28152, 3632, 3002, 508, 367, 1304, 304, 4505, 341, 1475, 705, 11356, 29889, 2, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, 319, 1799, 9047, 13566, 29901, 1763, 2334, 263, 341, 1475, 705, 11356, 899, 369, 393, 3913, 21531, 20159, 297, 5132, 29892, 366, 723, 817, 304, 437, 278, 1494, 29901, 13, 13, 29896, 29889, 16052, 263, 5132, 3577, 363, 21531, 20159, 29889, 1670, 526, 3196, 3987, 3625, 29892, 1316, 408, 402, 566, 2918, 29892, 360, 291, 952, 375, 29892, 470, 4560, 7354, 29899, 29873, 1388, 29889, 4525, 9741, 3867, 14009, 322, 848, 12286, 363, 20602, 25002, 15939, 1934, 29892, 1316, 408, 3632, 327, 2270, 6471, 29892, 350, 9890, 3694, 29892, 470, 28152, 3632, 3002, 29889, 13, 29906, 29889, 16314, 278, 341, 1475, 705, 11356, 3748, 7613, 408, 263, 25002, 2913, 29889, 910, 1033, 367, 2309, 773, 263, 848, 3829, 4944, 491, 278, 21531, 20159, 3577, 29892, 1316, 408, 263, 3053, 506, 616, 4280, 470, 263, 478, 2035, 29367, 29899, 29934, 4512, 4280, 29889, 7806, 6862, 373, 278, 7613, 723, 3928, 304, 263, 1298, 297, 278, 2913, 29892, 322, 278, 29086, 373, 278, 7613, 723, 367, 9875, 408, 376, 5391, 267, 29908, 297, 278, 2913, 29889, 13, 29941, 29889, 11796, 29872, 278, 25002, 15939, 1934, 310, 278, 2913, 29889, 910, 1033, 367, 2309, 773, 14009, 4944, 491, 278, 21531, 20159, 3577, 29889, 1152, 1342, 29892, 366, 1033, 10272, 278, 28152, 3632, 3002, 310, 278, 2913, 773, 278, 28152, 3632, 3002, 5687, 29889, 13, 29946, 29889, 4803, 278, 2472, 515, 278, 25002, 15939, 1934, 304, 8161, 278, 14354, 310, 278, 29086, 373, 278, 7613, 29889, 910, 1033, 367, 2309, 773, 263, 10296, 310, 19475, 24481, 322, 8720, 29889, 1152, 1342, 29892, 565, 278, 28152, 3632, 3002, 14088, 393, 727, 526, 3023, 26532, 297, 278, 2913, 29892, 21050, 297, 263, 6862, 8267, 29892, 769, 366, 1033, 671, 263, 8424, 310, 775, 304, 10115, 393, 278, 29086, 526, 5982, 297, 278, 3023, 26995, 310, 278, 7613, 29889, 13, 29945, 29889, 4803, 445, 2472, 304, 4505, 278, 341, 1475, 705, 11356, 3748, 29889, 9038, 366, 1073, 278, 14354, 310, 278, 29086, 29892, 366, 1033, 671, 263, 8424, 310, 775, 304, 6336, 2828, 373, 278, 9109, 25256, 322, 2821, 278, 7613, 29889, 13, 13, 2776, 3236, 29892, 445, 338, 925, 697, 1950, 2948, 304, 16049, 263, 341, 1475, 705, 11356, 899, 369, 393, 3913, 21531, 20159, 297, 5132, 29889, 1670, 1122, 367, 916, 5837, 304, 437, 372, 29892, 8679, 373, 278, 2702, 4902, 310, 278, 5314, 29889, 2398, 29892, 445, 881, 2367, 366, 263, 2498, 2969, 310, 920, 304, 679, 4687, 411, 1316, 263, 2060, 29889, 2]} +{"input_ids": [1, 319, 13563, 1546, 263, 12758, 1404, 322, 385, 23116, 21082, 20255, 29889, 450, 20255, 4076, 8444, 29892, 13173, 29892, 322, 1248, 568, 6089, 304, 278, 1404, 29915, 29879, 5155, 29889, 29871, 3148, 1001, 29901, 920, 1033, 474, 2334, 263, 29086, 705, 11356, 5687, 393, 3667, 4637, 21531, 20159, 304, 4505, 1045, 3163, 29973, 29871, 319, 1799, 9047, 13566, 29901, 739, 29915, 29879, 1950, 304, 671, 21531, 20159, 304, 4505, 341, 1475, 705, 11356, 29892, 541, 372, 723, 5517, 367, 3755, 4280, 29889, 838, 5869, 293, 20159, 338, 263, 5443, 310, 23964, 393, 11898, 278, 4426, 310, 8162, 393, 526, 21634, 1090, 9126, 316, 689, 800, 29892, 1316, 408, 16116, 292, 322, 289, 2548, 29889, 512, 278, 3030, 310, 341, 1475, 705, 11356, 29892, 445, 1033, 367, 1304, 304, 27599, 278, 3829, 310, 278, 3748, 7613, 322, 8161, 278, 14354, 310, 29086, 2729, 373, 3058, 25002, 15939, 1934, 29889, 2398, 29892, 16049, 445, 723, 1996, 263, 6483, 8004, 310, 1716, 21531, 20159, 322, 278, 3748, 310, 341, 1475, 705, 11356, 29889, 739, 29915, 29879, 451, 1554, 393, 508, 367, 10824, 297, 263, 2560, 1234, 21106, 29879, 29958, 2, 3148, 1001, 29901, 1033, 366, 2367, 592, 278, 330, 391, 310, 920, 372, 1033, 2466, 29973, 29871, 319, 1799, 9047, 13566, 29901, 18585, 29889, 450, 6996, 2969, 723, 367, 304, 671, 21531, 20159, 304, 27599, 278, 3829, 310, 278, 341, 1475, 705, 11356, 3748, 7613, 322, 8161, 278, 14354, 310, 29086, 2729, 373, 3058, 25002, 15939, 1934, 29889, 910, 1033, 25135, 15783, 278, 3748, 7613, 408, 263, 25002, 2913, 322, 773, 25002, 15939, 1934, 1316, 408, 3632, 327, 2270, 6471, 322, 350, 9890, 3694, 304, 10115, 2472, 1048, 278, 4423, 310, 29086, 373, 278, 7613, 29889, 2398, 29892, 408, 306, 5276, 1434, 29892, 16049, 445, 723, 1996, 263, 6483, 8004, 310, 1716, 21531, 20159, 322, 278, 3748, 310, 341, 1475, 705, 11356, 29892, 577, 372, 29915, 29879, 451, 1554, 393, 508, 367, 10824, 297, 263, 2560, 1234, 21106, 29879, 29958, 2, 3148, 1001, 29901, 2367, 592, 263, 2702, 1342, 29871, 319, 1799, 9047, 13566, 29901, 1094, 263, 2702, 1342, 29892, 7755, 591, 505, 263, 341, 1475, 705, 11356, 3748, 7613, 411, 278, 1494, 5285, 29901, 13, 28956, 13, 29871, 29896, 259, 29896, 259, 29896, 259, 29896, 259, 29896, 13, 29871, 29896, 29871, 1577, 29871, 1577, 29871, 1577, 259, 29896, 13, 29871, 29896, 29871, 1577, 29871, 1577, 29871, 1577, 259, 29896, 13, 29871, 29896, 29871, 1577, 29871, 1577, 29871, 1577, 259, 29896, 13, 29871, 29896, 259, 29896, 259, 29896, 259, 29896, 259, 29896, 13, 28956, 13, 797, 445, 1206, 29892, 278, 3748, 7613, 508, 367, 9875, 408, 263, 25002, 2913, 29892, 988, 1269, 6862, 373, 278, 7613, 16161, 304, 263, 1298, 297, 278, 2913, 29889, 450, 29086, 373, 278, 7613, 508, 367, 2714, 310, 408, 376, 5391, 267, 29908, 297, 278, 2913, 29892, 607, 526, 3291, 393, 526, 451, 760, 310, 278, 2913, 29889, 5293, 21531, 20159, 29892, 591, 508, 27599, 278, 3829, 310, 445, 2913, 322, 8161, 278, 14354, 310, 278, 29086, 2729, 373, 3058, 25002, 15939, 1934, 29889, 13, 13, 2831, 2777, 29892, 697, 982, 304, 437, 445, 338, 304, 10272, 278, 3632, 327, 2270, 6471, 310, 278, 2913, 29889, 15089, 327, 2270, 6471, 526, 263, 982, 310, 20766, 278, 26532, 297, 263, 2913, 29892, 322, 896, 508, 367, 1304, 304, 10115, 2472, 1048, 278, 14354, 310, 278, 29086, 373, 278, 341, 1475, 705, 11356, 7613, 29889, 512, 278, 1206, 310, 278, 7613, 2038, 29892, 591, 508, 10272, 278, 937, 3632, 327, 2270, 2318, 310, 278, 2913, 29892, 607, 16612, 278, 1353, 322, 376, 12181, 29908, 310, 278, 26532, 297, 278, 2913, 29889, 512, 445, 1206, 29892, 278, 937, 3632, 327, 2270, 2318, 723, 2649, 502, 393, 727, 526, 3023, 26532, 297, 278, 2913, 29892, 21050, 297, 263, 6862, 8267, 29889, 13, 13, 29933, 1463, 373, 445, 2472, 29892, 591, 508, 10115, 393, 278, 29086, 297, 278, 3748, 526, 5982, 297, 278, 3023, 26995, 310, 278, 7613, 29892, 408, 4318, 2400, 29901, 13, 28956, 13, 29871, 29896, 259, 29896, 259, 29896, 259, 29896, 259, 29896, 13, 29871, 29896, 29871, 341, 29871, 1577, 29871, 1577, 259, 29896, 13, 29871, 29896, 29871, 1577, 29871, 1577, 29871, 1577, 259, 29896, 13, 29871, 29896, 29871, 1577, 29871, 1577, 29871, 341, 259, 29896, 13, 29871, 29896, 259, 29896, 259, 29896, 259, 29896, 259, 29896, 13, 28956, 13, 2776, 3236, 29892, 445, 338, 925, 697, 1342, 310, 920, 21531, 20159, 1033, 367, 1304, 304, 4505, 341, 1475, 705, 11356, 29889, 1670, 526, 1784, 916, 1950, 13501, 29892, 322, 278, 2702, 4902, 310, 920, 304, 2334, 963, 723, 8839, 373, 278, 2702, 3748, 7613, 322, 278, 25002, 15939, 1934, 1641, 1304, 29889, 1094, 306, 5276, 1434, 29892, 16049, 263, 1650, 763, 445, 723, 1996, 263, 6483, 8004, 310, 1716, 21531, 20159, 322, 278, 3748, 310, 341, 1475, 705, 11356, 21106, 29879, 29958, 2, 3148, 1001, 29901, 10241, 474, 471, 4856, 411, 263, 6483, 8004, 310, 1716, 29892, 1033, 366, 5649, 278, 1021, 6964, 304, 592, 411, 393, 11833, 29973, 29871, 319, 1799, 9047, 13566, 29901, 18585, 29889, 17090, 366, 505, 263, 6483, 8004, 310, 1716, 21531, 20159, 322, 278, 3748, 310, 341, 1475, 705, 11356, 29892, 1244, 338, 920, 366, 1795, 671, 21531, 20159, 304, 4505, 263, 341, 1475, 705, 11356, 3748, 7613, 29901, 13, 13, 29896, 29889, 16314, 278, 341, 1475, 705, 11356, 3748, 7613, 408, 263, 25002, 2913, 29889, 7806, 6862, 373, 278, 7613, 16161, 304, 263, 1298, 297, 278, 2913, 29892, 322, 278, 29086, 373, 278, 7613, 508, 367, 2714, 310, 408, 376, 5391, 267, 29908, 297, 278, 2913, 29892, 607, 526, 3291, 393, 526, 451, 760, 310, 278, 2913, 29889, 13, 29906, 29889, 4803, 25002, 15939, 1934, 304, 10115, 2472, 1048, 278, 14354, 310, 278, 29086, 373, 278, 7613, 29889, 1670, 526, 1784, 1950, 25002, 15939, 1934, 393, 1033, 367, 1304, 363, 445, 29892, 1316, 408, 3632, 327, 2270, 6471, 29892, 350, 9890, 3694, 29892, 470, 28152, 3632, 3002, 29889, 13, 29941, 29889, 4803, 278, 2472, 10115, 1127, 515, 278, 25002, 15939, 1934, 304, 8161, 278, 14354, 310, 278, 29086, 373, 278, 7613, 29889, 1152, 1342, 29892, 565, 278, 937, 3632, 327, 2270, 2318, 310, 278, 2913, 14088, 393, 727, 526, 3023, 26532, 297, 278, 2913, 29892, 21050, 297, 263, 6862, 8267, 29892, 769, 591, 508, 10115, 393, 278, 29086, 526, 5982, 297, 278, 3023, 26995, 310, 278, 7613, 29889, 13, 29946, 29889, 4803, 445, 2472, 304, 4505, 278, 341, 1475, 705, 11356, 3748, 29889, 9038, 366, 1073, 278, 14354, 310, 278, 29086, 29892, 366, 508, 671, 445, 2472, 304, 4772, 963, 322, 2821, 278, 7613, 29889, 13, 13, 2776, 3236, 29892, 445, 338, 925, 697, 1950, 2948, 304, 773, 21531, 20159, 304, 4505, 341, 1475, 705, 11356, 29889, 1670, 1122, 367, 916, 5837, 304, 437, 372, 29892, 8679, 373, 278, 2702, 3748, 7613, 322, 278, 25002, 15939, 1934, 1641, 1304, 29889, 2398, 29892, 445, 881, 2367, 366, 263, 2498, 2969, 310, 920, 21531, 20159, 508, 367, 1304, 304, 4505, 341, 1475, 705, 11356, 21106, 29879, 29958, 2, 3148, 1001, 29901, 920, 723, 474, 437, 372, 411, 28152, 3632, 3002, 470, 738, 310, 278, 916, 25002, 15939, 1934, 29973, 29871, 319, 1799, 9047, 13566, 29901, 1763, 671, 28152, 3632, 3002, 470, 738, 310, 278, 916, 25002, 15939, 1934, 304, 4505, 263, 341, 1475, 705, 11356, 3748, 7613, 29892, 366, 723, 1101, 278, 1021, 6996, 6576, 408, 5439, 2038, 29892, 541, 411, 777, 12651, 297, 278, 2702, 4902, 29889, 13, 13, 2831, 2777, 29892, 1244, 338, 920, 366, 1795, 671, 28152, 3632, 3002, 304, 4505, 263, 341, 1475, 705, 11356, 3748, 7613, 29901, 13, 13, 29896, 29889, 16314, 278, 341, 1475, 705, 11356, 3748, 7613, 408, 263, 25002, 2913, 29889, 7806, 6862, 373, 278, 7613, 16161, 304, 263, 1298, 297, 278, 2913, 29892, 322, 278, 29086, 373, 278, 7613, 508, 367, 2714, 310, 408, 376, 5391, 267, 29908, 297, 278, 2913, 29892, 607, 526, 3291, 393, 526, 451, 760, 310, 278, 2913, 29889, 13, 29906, 29889, 11796, 29872, 278, 28152, 3632, 3002, 310, 278, 2913, 29889, 9034, 9696, 3632, 3002, 338, 263, 982, 310, 20766, 278, 26532, 297, 263, 2913, 975, 931, 29892, 322, 372, 508, 367, 1304, 304, 10115, 2472, 1048, 278, 14354, 310, 278, 29086, 373, 278, 341, 1475, 705, 11356, 7613, 29889, 13, 29941, 29889, 4803, 278, 2472, 515, 278, 28152, 3632, 3002, 304, 8161, 278, 14354, 310, 278, 29086, 373, 278, 7613, 29889, 1152, 1342, 29892, 565, 278, 28152, 3632, 3002, 14088, 393, 727, 526, 3023, 26532, 297, 278, 2913, 29892, 21050, 297, 263, 6862, 8267, 29892, 769, 591, 508, 10115, 393, 278, 29086, 526, 5982, 297, 278, 3023, 26995, 310, 278, 7613, 29889, 13, 29946, 29889, 4803, 445, 2472, 304, 4505, 278, 341, 1475, 705, 11356, 3748, 29889, 9038, 366, 1073, 278, 14354, 310, 278, 29086, 29892, 366, 508, 671, 445, 2472, 304, 4772, 963, 322, 2821, 278, 7613, 29889, 13, 13, 2776, 3236, 29892, 445, 338, 925, 697, 1950, 2948, 304, 773, 28152, 3632, 3002, 304, 4505, 341, 1475, 705, 11356, 29889, 1670, 1122, 367, 916, 5837, 304, 437, 372, 29892, 8679, 373, 278, 2702, 3748, 7613, 322, 278, 2702, 4902, 310, 278, 28152, 3632, 3002, 16287, 29889, 2398, 29892, 445, 881, 2367, 366, 263, 2498, 2969, 310, 920, 28152, 3632, 3002, 508, 367, 1304, 304, 4505, 341, 1475, 705, 11356, 21106, 29879, 29958, 2, 3148, 1001, 29901, 920, 1033, 474, 2334, 445, 297, 3017, 29973, 29871, 319, 1799, 9047, 13566, 29901, 1763, 2334, 263, 341, 1475, 705, 11356, 899, 369, 393, 3913, 21531, 20159, 297, 5132, 29892, 366, 723, 817, 304, 437, 278, 1494, 29901, 13, 13, 29896, 29889, 16052, 263, 5132, 3577, 363, 21531, 20159, 29889, 1670, 526, 3196, 3987, 3625, 29892, 1316, 408, 402, 566, 2918, 29892, 360, 291, 952, 375, 29892, 470, 4560, 7354, 29899, 29873, 1388, 29889, 4525, 9741, 3867, 14009, 322, 848, 12286, 363, 20602, 25002, 15939, 1934, 29892, 1316, 408, 3632, 327, 2270, 6471, 29892, 350, 9890, 3694, 29892, 470, 28152, 3632, 3002, 29889, 13, 29906, 29889, 16314, 278, 341, 1475, 705, 11356, 3748, 7613, 408, 263, 25002, 2913, 29889, 910, 1033, 367, 2309, 773, 263, 848, 3829, 4944, 491, 278, 21531, 20159, 3577, 29892, 1316, 408, 263, 3053, 506, 616, 4280, 470, 263, 478, 2035, 29367, 29899, 29934, 4512, 4280, 29889, 7806, 6862, 373, 278, 7613, 723, 3928, 304, 263, 1298, 297, 278, 2913, 29892, 322, 278, 29086, 373, 278, 7613, 723, 367, 9875, 408, 376, 5391, 267, 29908, 297, 278, 2913, 29889, 13, 29941, 29889, 11796, 29872, 278, 25002, 15939, 1934, 310, 278, 2913, 29889, 910, 1033, 367, 2309, 773, 14009, 4944, 491, 278, 21531, 20159, 3577, 29889, 1152, 1342, 29892, 366, 1033, 10272, 278, 28152, 3632, 3002, 310, 278, 2913, 773, 278, 28152, 3632, 3002, 5687, 29889, 13, 29946, 29889, 4803, 278, 2472, 515, 278, 25002, 15939, 1934, 304, 8161, 278, 14354, 310, 278, 29086, 373, 278, 7613, 29889, 910, 1033, 367, 2309, 773, 263, 10296, 310, 19475, 24481, 322, 8720, 29889, 1152, 1342, 29892, 565, 278, 28152, 3632, 3002, 14088, 393, 727, 526, 3023, 26532, 297, 278, 2913, 29892, 21050, 297, 263, 6862, 8267, 29892, 769, 366, 1033, 671, 263, 8424, 310, 775, 304, 10115, 393, 278, 29086, 526, 5982, 297, 278, 3023, 26995, 310, 278, 7613, 29889, 13, 29945, 29889, 4803, 445, 2472, 304, 4505, 278, 341, 1475, 705, 11356, 3748, 29889, 9038, 366, 1073, 278, 14354, 310, 278, 29086, 29892, 366, 1033, 671, 263, 8424, 310, 775, 304, 6336, 2828, 373, 278, 9109, 25256, 322, 2821, 278, 7613, 29889, 13, 13, 2776, 3236, 29892, 445, 338, 925, 697, 1950, 2948, 304, 16049, 263, 341, 1475, 705, 11356, 899, 369, 393, 3913, 21531, 20159, 297, 5132, 29889, 1670, 1122, 367, 916, 5837, 304, 437, 372, 29892, 8679, 373, 278, 2702, 4902, 310, 278, 5314, 29889, 2398, 29892, 445, 881, 2367, 366, 263, 2498, 2969, 310, 920, 304, 679, 4687, 411, 1316, 263, 2060, 21106, 29879, 29958, 2], "attention_mask": [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], "labels": [-100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, 319, 1799, 9047, 13566, 29901, 739, 29915, 29879, 1950, 304, 671, 21531, 20159, 304, 4505, 341, 1475, 705, 11356, 29892, 541, 372, 723, 5517, 367, 3755, 4280, 29889, 838, 5869, 293, 20159, 338, 263, 5443, 310, 23964, 393, 11898, 278, 4426, 310, 8162, 393, 526, 21634, 1090, 9126, 316, 689, 800, 29892, 1316, 408, 16116, 292, 322, 289, 2548, 29889, 512, 278, 3030, 310, 341, 1475, 705, 11356, 29892, 445, 1033, 367, 1304, 304, 27599, 278, 3829, 310, 278, 3748, 7613, 322, 8161, 278, 14354, 310, 29086, 2729, 373, 3058, 25002, 15939, 1934, 29889, 2398, 29892, 16049, 445, 723, 1996, 263, 6483, 8004, 310, 1716, 21531, 20159, 322, 278, 3748, 310, 341, 1475, 705, 11356, 29889, 739, 29915, 29879, 451, 1554, 393, 508, 367, 10824, 297, 263, 2560, 1234, 21106, 29879, 29958, 2, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, 319, 1799, 9047, 13566, 29901, 18585, 29889, 450, 6996, 2969, 723, 367, 304, 671, 21531, 20159, 304, 27599, 278, 3829, 310, 278, 341, 1475, 705, 11356, 3748, 7613, 322, 8161, 278, 14354, 310, 29086, 2729, 373, 3058, 25002, 15939, 1934, 29889, 910, 1033, 25135, 15783, 278, 3748, 7613, 408, 263, 25002, 2913, 322, 773, 25002, 15939, 1934, 1316, 408, 3632, 327, 2270, 6471, 322, 350, 9890, 3694, 304, 10115, 2472, 1048, 278, 4423, 310, 29086, 373, 278, 7613, 29889, 2398, 29892, 408, 306, 5276, 1434, 29892, 16049, 445, 723, 1996, 263, 6483, 8004, 310, 1716, 21531, 20159, 322, 278, 3748, 310, 341, 1475, 705, 11356, 29892, 577, 372, 29915, 29879, 451, 1554, 393, 508, 367, 10824, 297, 263, 2560, 1234, 21106, 29879, 29958, 2, -100, -100, -100, -100, -100, -100, -100, -100, -100, 319, 1799, 9047, 13566, 29901, 1094, 263, 2702, 1342, 29892, 7755, 591, 505, 263, 341, 1475, 705, 11356, 3748, 7613, 411, 278, 1494, 5285, 29901, 13, 28956, 13, 29871, 29896, 259, 29896, 259, 29896, 259, 29896, 259, 29896, 13, 29871, 29896, 29871, 1577, 29871, 1577, 29871, 1577, 259, 29896, 13, 29871, 29896, 29871, 1577, 29871, 1577, 29871, 1577, 259, 29896, 13, 29871, 29896, 29871, 1577, 29871, 1577, 29871, 1577, 259, 29896, 13, 29871, 29896, 259, 29896, 259, 29896, 259, 29896, 259, 29896, 13, 28956, 13, 797, 445, 1206, 29892, 278, 3748, 7613, 508, 367, 9875, 408, 263, 25002, 2913, 29892, 988, 1269, 6862, 373, 278, 7613, 16161, 304, 263, 1298, 297, 278, 2913, 29889, 450, 29086, 373, 278, 7613, 508, 367, 2714, 310, 408, 376, 5391, 267, 29908, 297, 278, 2913, 29892, 607, 526, 3291, 393, 526, 451, 760, 310, 278, 2913, 29889, 5293, 21531, 20159, 29892, 591, 508, 27599, 278, 3829, 310, 445, 2913, 322, 8161, 278, 14354, 310, 278, 29086, 2729, 373, 3058, 25002, 15939, 1934, 29889, 13, 13, 2831, 2777, 29892, 697, 982, 304, 437, 445, 338, 304, 10272, 278, 3632, 327, 2270, 6471, 310, 278, 2913, 29889, 15089, 327, 2270, 6471, 526, 263, 982, 310, 20766, 278, 26532, 297, 263, 2913, 29892, 322, 896, 508, 367, 1304, 304, 10115, 2472, 1048, 278, 14354, 310, 278, 29086, 373, 278, 341, 1475, 705, 11356, 7613, 29889, 512, 278, 1206, 310, 278, 7613, 2038, 29892, 591, 508, 10272, 278, 937, 3632, 327, 2270, 2318, 310, 278, 2913, 29892, 607, 16612, 278, 1353, 322, 376, 12181, 29908, 310, 278, 26532, 297, 278, 2913, 29889, 512, 445, 1206, 29892, 278, 937, 3632, 327, 2270, 2318, 723, 2649, 502, 393, 727, 526, 3023, 26532, 297, 278, 2913, 29892, 21050, 297, 263, 6862, 8267, 29889, 13, 13, 29933, 1463, 373, 445, 2472, 29892, 591, 508, 10115, 393, 278, 29086, 297, 278, 3748, 526, 5982, 297, 278, 3023, 26995, 310, 278, 7613, 29892, 408, 4318, 2400, 29901, 13, 28956, 13, 29871, 29896, 259, 29896, 259, 29896, 259, 29896, 259, 29896, 13, 29871, 29896, 29871, 341, 29871, 1577, 29871, 1577, 259, 29896, 13, 29871, 29896, 29871, 1577, 29871, 1577, 29871, 1577, 259, 29896, 13, 29871, 29896, 29871, 1577, 29871, 1577, 29871, 341, 259, 29896, 13, 29871, 29896, 259, 29896, 259, 29896, 259, 29896, 259, 29896, 13, 28956, 13, 2776, 3236, 29892, 445, 338, 925, 697, 1342, 310, 920, 21531, 20159, 1033, 367, 1304, 304, 4505, 341, 1475, 705, 11356, 29889, 1670, 526, 1784, 916, 1950, 13501, 29892, 322, 278, 2702, 4902, 310, 920, 304, 2334, 963, 723, 8839, 373, 278, 2702, 3748, 7613, 322, 278, 25002, 15939, 1934, 1641, 1304, 29889, 1094, 306, 5276, 1434, 29892, 16049, 263, 1650, 763, 445, 723, 1996, 263, 6483, 8004, 310, 1716, 21531, 20159, 322, 278, 3748, 310, 341, 1475, 705, 11356, 21106, 29879, 29958, 2, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, 319, 1799, 9047, 13566, 29901, 18585, 29889, 17090, 366, 505, 263, 6483, 8004, 310, 1716, 21531, 20159, 322, 278, 3748, 310, 341, 1475, 705, 11356, 29892, 1244, 338, 920, 366, 1795, 671, 21531, 20159, 304, 4505, 263, 341, 1475, 705, 11356, 3748, 7613, 29901, 13, 13, 29896, 29889, 16314, 278, 341, 1475, 705, 11356, 3748, 7613, 408, 263, 25002, 2913, 29889, 7806, 6862, 373, 278, 7613, 16161, 304, 263, 1298, 297, 278, 2913, 29892, 322, 278, 29086, 373, 278, 7613, 508, 367, 2714, 310, 408, 376, 5391, 267, 29908, 297, 278, 2913, 29892, 607, 526, 3291, 393, 526, 451, 760, 310, 278, 2913, 29889, 13, 29906, 29889, 4803, 25002, 15939, 1934, 304, 10115, 2472, 1048, 278, 14354, 310, 278, 29086, 373, 278, 7613, 29889, 1670, 526, 1784, 1950, 25002, 15939, 1934, 393, 1033, 367, 1304, 363, 445, 29892, 1316, 408, 3632, 327, 2270, 6471, 29892, 350, 9890, 3694, 29892, 470, 28152, 3632, 3002, 29889, 13, 29941, 29889, 4803, 278, 2472, 10115, 1127, 515, 278, 25002, 15939, 1934, 304, 8161, 278, 14354, 310, 278, 29086, 373, 278, 7613, 29889, 1152, 1342, 29892, 565, 278, 937, 3632, 327, 2270, 2318, 310, 278, 2913, 14088, 393, 727, 526, 3023, 26532, 297, 278, 2913, 29892, 21050, 297, 263, 6862, 8267, 29892, 769, 591, 508, 10115, 393, 278, 29086, 526, 5982, 297, 278, 3023, 26995, 310, 278, 7613, 29889, 13, 29946, 29889, 4803, 445, 2472, 304, 4505, 278, 341, 1475, 705, 11356, 3748, 29889, 9038, 366, 1073, 278, 14354, 310, 278, 29086, 29892, 366, 508, 671, 445, 2472, 304, 4772, 963, 322, 2821, 278, 7613, 29889, 13, 13, 2776, 3236, 29892, 445, 338, 925, 697, 1950, 2948, 304, 773, 21531, 20159, 304, 4505, 341, 1475, 705, 11356, 29889, 1670, 1122, 367, 916, 5837, 304, 437, 372, 29892, 8679, 373, 278, 2702, 3748, 7613, 322, 278, 25002, 15939, 1934, 1641, 1304, 29889, 2398, 29892, 445, 881, 2367, 366, 263, 2498, 2969, 310, 920, 21531, 20159, 508, 367, 1304, 304, 4505, 341, 1475, 705, 11356, 21106, 29879, 29958, 2, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, 319, 1799, 9047, 13566, 29901, 1763, 671, 28152, 3632, 3002, 470, 738, 310, 278, 916, 25002, 15939, 1934, 304, 4505, 263, 341, 1475, 705, 11356, 3748, 7613, 29892, 366, 723, 1101, 278, 1021, 6996, 6576, 408, 5439, 2038, 29892, 541, 411, 777, 12651, 297, 278, 2702, 4902, 29889, 13, 13, 2831, 2777, 29892, 1244, 338, 920, 366, 1795, 671, 28152, 3632, 3002, 304, 4505, 263, 341, 1475, 705, 11356, 3748, 7613, 29901, 13, 13, 29896, 29889, 16314, 278, 341, 1475, 705, 11356, 3748, 7613, 408, 263, 25002, 2913, 29889, 7806, 6862, 373, 278, 7613, 16161, 304, 263, 1298, 297, 278, 2913, 29892, 322, 278, 29086, 373, 278, 7613, 508, 367, 2714, 310, 408, 376, 5391, 267, 29908, 297, 278, 2913, 29892, 607, 526, 3291, 393, 526, 451, 760, 310, 278, 2913, 29889, 13, 29906, 29889, 11796, 29872, 278, 28152, 3632, 3002, 310, 278, 2913, 29889, 9034, 9696, 3632, 3002, 338, 263, 982, 310, 20766, 278, 26532, 297, 263, 2913, 975, 931, 29892, 322, 372, 508, 367, 1304, 304, 10115, 2472, 1048, 278, 14354, 310, 278, 29086, 373, 278, 341, 1475, 705, 11356, 7613, 29889, 13, 29941, 29889, 4803, 278, 2472, 515, 278, 28152, 3632, 3002, 304, 8161, 278, 14354, 310, 278, 29086, 373, 278, 7613, 29889, 1152, 1342, 29892, 565, 278, 28152, 3632, 3002, 14088, 393, 727, 526, 3023, 26532, 297, 278, 2913, 29892, 21050, 297, 263, 6862, 8267, 29892, 769, 591, 508, 10115, 393, 278, 29086, 526, 5982, 297, 278, 3023, 26995, 310, 278, 7613, 29889, 13, 29946, 29889, 4803, 445, 2472, 304, 4505, 278, 341, 1475, 705, 11356, 3748, 29889, 9038, 366, 1073, 278, 14354, 310, 278, 29086, 29892, 366, 508, 671, 445, 2472, 304, 4772, 963, 322, 2821, 278, 7613, 29889, 13, 13, 2776, 3236, 29892, 445, 338, 925, 697, 1950, 2948, 304, 773, 28152, 3632, 3002, 304, 4505, 341, 1475, 705, 11356, 29889, 1670, 1122, 367, 916, 5837, 304, 437, 372, 29892, 8679, 373, 278, 2702, 3748, 7613, 322, 278, 2702, 4902, 310, 278, 28152, 3632, 3002, 16287, 29889, 2398, 29892, 445, 881, 2367, 366, 263, 2498, 2969, 310, 920, 28152, 3632, 3002, 508, 367, 1304, 304, 4505, 341, 1475, 705, 11356, 21106, 29879, 29958, 2, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, 319, 1799, 9047, 13566, 29901, 1763, 2334, 263, 341, 1475, 705, 11356, 899, 369, 393, 3913, 21531, 20159, 297, 5132, 29892, 366, 723, 817, 304, 437, 278, 1494, 29901, 13, 13, 29896, 29889, 16052, 263, 5132, 3577, 363, 21531, 20159, 29889, 1670, 526, 3196, 3987, 3625, 29892, 1316, 408, 402, 566, 2918, 29892, 360, 291, 952, 375, 29892, 470, 4560, 7354, 29899, 29873, 1388, 29889, 4525, 9741, 3867, 14009, 322, 848, 12286, 363, 20602, 25002, 15939, 1934, 29892, 1316, 408, 3632, 327, 2270, 6471, 29892, 350, 9890, 3694, 29892, 470, 28152, 3632, 3002, 29889, 13, 29906, 29889, 16314, 278, 341, 1475, 705, 11356, 3748, 7613, 408, 263, 25002, 2913, 29889, 910, 1033, 367, 2309, 773, 263, 848, 3829, 4944, 491, 278, 21531, 20159, 3577, 29892, 1316, 408, 263, 3053, 506, 616, 4280, 470, 263, 478, 2035, 29367, 29899, 29934, 4512, 4280, 29889, 7806, 6862, 373, 278, 7613, 723, 3928, 304, 263, 1298, 297, 278, 2913, 29892, 322, 278, 29086, 373, 278, 7613, 723, 367, 9875, 408, 376, 5391, 267, 29908, 297, 278, 2913, 29889, 13, 29941, 29889, 11796, 29872, 278, 25002, 15939, 1934, 310, 278, 2913, 29889, 910, 1033, 367, 2309, 773, 14009, 4944, 491, 278, 21531, 20159, 3577, 29889, 1152, 1342, 29892, 366, 1033, 10272, 278, 28152, 3632, 3002, 310, 278, 2913, 773, 278, 28152, 3632, 3002, 5687, 29889, 13, 29946, 29889, 4803, 278, 2472, 515, 278, 25002, 15939, 1934, 304, 8161, 278, 14354, 310, 278, 29086, 373, 278, 7613, 29889, 910, 1033, 367, 2309, 773, 263, 10296, 310, 19475, 24481, 322, 8720, 29889, 1152, 1342, 29892, 565, 278, 28152, 3632, 3002, 14088, 393, 727, 526, 3023, 26532, 297, 278, 2913, 29892, 21050, 297, 263, 6862, 8267, 29892, 769, 366, 1033, 671, 263, 8424, 310, 775, 304, 10115, 393, 278, 29086, 526, 5982, 297, 278, 3023, 26995, 310, 278, 7613, 29889, 13, 29945, 29889, 4803, 445, 2472, 304, 4505, 278, 341, 1475, 705, 11356, 3748, 29889, 9038, 366, 1073, 278, 14354, 310, 278, 29086, 29892, 366, 1033, 671, 263, 8424, 310, 775, 304, 6336, 2828, 373, 278, 9109, 25256, 322, 2821, 278, 7613, 29889, 13, 13, 2776, 3236, 29892, 445, 338, 925, 697, 1950, 2948, 304, 16049, 263, 341, 1475, 705, 11356, 899, 369, 393, 3913, 21531, 20159, 297, 5132, 29889, 1670, 1122, 367, 916, 5837, 304, 437, 372, 29892, 8679, 373, 278, 2702, 4902, 310, 278, 5314, 29889, 2398, 29892, 445, 881, 2367, 366, 263, 2498, 2969, 310, 920, 304, 679, 4687, 411, 1316, 263, 2060, 21106, 29879, 29958, 2]} From 51682d6e13a5caaef8a13f43ffccee8d846573fc Mon Sep 17 00:00:00 2001 From: Wing Lian Date: Mon, 25 Sep 2023 13:14:40 -0400 Subject: [PATCH 07/16] use a new prompter and support fastchat conversation type --- .../prompt_strategies/sharegpt_jokes.py | 4 +-- .../prompt_strategies/sharegpt_simple.py | 8 ++--- src/axolotl/prompters.py | 30 ++++++++----------- src/axolotl/utils/data.py | 4 +-- tests/test_prompt_tokenizers.py | 6 ++-- 5 files changed, 24 insertions(+), 28 deletions(-) diff --git a/src/axolotl/prompt_strategies/sharegpt_jokes.py b/src/axolotl/prompt_strategies/sharegpt_jokes.py index ac424bf7c7..404302c81e 100644 --- a/src/axolotl/prompt_strategies/sharegpt_jokes.py +++ b/src/axolotl/prompt_strategies/sharegpt_jokes.py @@ -1,11 +1,11 @@ """Module for Jokes prompts using sharegpt style """ from axolotl.prompt_tokenizers import ShareGPTPromptTokenizingStrategy -from axolotl.prompters import PromptStyle, ShareGPTPrompter +from axolotl.prompters import ShareGPTPrompterV2 def load(tokenizer, cfg): return SimpleJokesShareGPTPromptTokenizingStrategy( - ShareGPTPrompter(PromptStyle.CHAT.value), + ShareGPTPrompterV2(), tokenizer, cfg.train_on_inputs, cfg.sequence_len, diff --git a/src/axolotl/prompt_strategies/sharegpt_simple.py b/src/axolotl/prompt_strategies/sharegpt_simple.py index bfe0d164bc..a12ab1eab9 100644 --- a/src/axolotl/prompt_strategies/sharegpt_simple.py +++ b/src/axolotl/prompt_strategies/sharegpt_simple.py @@ -1,12 +1,12 @@ """Module containing the SimpleShareGPTPromptTokenizingStrategy class""" from axolotl.prompt_tokenizers import ShareGPTPromptTokenizingStrategy -from axolotl.prompters import PromptStyle, ShareGPTPrompter +from axolotl.prompters import ShareGPTPrompterV2 def load(tokenizer, cfg): return SimpleShareGPTPromptTokenizingStrategy( - ShareGPTPrompter(PromptStyle.CHAT.value), + ShareGPTPrompterV2(), tokenizer, cfg.train_on_inputs, cfg.sequence_len, @@ -15,7 +15,7 @@ def load(tokenizer, cfg): def load_role(tokenizer, cfg): return SimpleRoleShareGPTPromptTokenizingStrategy( - ShareGPTPrompter(PromptStyle.CHAT.value), + ShareGPTPrompterV2(), tokenizer, cfg.train_on_inputs, cfg.sequence_len, @@ -24,7 +24,7 @@ def load_role(tokenizer, cfg): def load_guanaco(tokenizer, cfg): return GuanacoShareGPTPromptTokenizingStrategy( - ShareGPTPrompter(PromptStyle.CHAT.value), + ShareGPTPrompterV2(), tokenizer, cfg.train_on_inputs, cfg.sequence_len, diff --git a/src/axolotl/prompters.py b/src/axolotl/prompters.py index afb1bb4a1c..5b955cc0b9 100644 --- a/src/axolotl/prompters.py +++ b/src/axolotl/prompters.py @@ -227,22 +227,9 @@ class ShareGPTPrompter: # pylint: disable=too-few-public-methods def __init__( self, - prompt_style=None, - system_prompt: Optional[str] = None, + prompt_style=None, # pylint: disable=unused-argument conversation: Optional[Union[str, Conversation]] = None, ): - if prompt_style != PromptStyle.CHAT.value: - raise ValueError( - f"unsupported prompt_style for ShareGPTPrompter({prompt_style})" - ) - system: str = ( - system_prompt - if system_prompt is not None - else ( - "A chat between a curious user and an artificial intelligence assistant. " - "The assistant gives helpful, detailed, and polite answers to the user's questions." - ) - ) if conversation: if isinstance(conversation, Conversation): self._conversation = conversation @@ -251,9 +238,6 @@ def __init__( else: self._conversation = get_conv_template("vicuna_v1.1") - if system: - self._conversation.system_message = system - def build_prompt(self, source) -> Generator[str, None, None]: if len(source) < 2: # If there isn't a back and forth conversation, ignore it @@ -293,3 +277,15 @@ def build_prompt(self, source) -> Generator[str, None, None]: if part[0] and not part[1]: LOG.warning(f"role with empty message: {part[0]}") yield part + + +class ShareGPTPrompterV2(ShareGPTPrompter): + """ + A V2 prompter that generates prompts for the ShareGPT + """ + + def __init__( + self, + conversation: Optional[Union[str, Conversation]] = None, + ): + super().__init__(conversation=conversation) diff --git a/src/axolotl/utils/data.py b/src/axolotl/utils/data.py index 9792371c7c..461067b4ec 100644 --- a/src/axolotl/utils/data.py +++ b/src/axolotl/utils/data.py @@ -35,7 +35,7 @@ MultipleChoiceConcisePrompter, MultipleChoiceExplainPrompter, ReflectAlpacaPrompter, - ShareGPTPrompter, + ShareGPTPrompterV2, SummarizeTLDRPrompter, ) from axolotl.utils.dict import DictDefault @@ -322,7 +322,7 @@ def for_d_in_datasets(dataset_configs): datasets.append(ds_wrapper) elif d_base_type == "sharegpt": ds_strategy = ShareGPTPromptTokenizingStrategy( - ShareGPTPrompter(d_prompt_style), + ShareGPTPrompterV2(d.conversation), tokenizer, cfg.train_on_inputs, cfg.sequence_len, diff --git a/tests/test_prompt_tokenizers.py b/tests/test_prompt_tokenizers.py index a455d8b513..9ee5ea588d 100644 --- a/tests/test_prompt_tokenizers.py +++ b/tests/test_prompt_tokenizers.py @@ -21,7 +21,7 @@ AlpacaPromptTokenizingStrategy, ShareGPTPromptTokenizingStrategy, ) -from axolotl.prompters import AlpacaPrompter, PromptStyle, ShareGPTPrompter +from axolotl.prompters import AlpacaPrompter, PromptStyle, ShareGPTPrompterV2 LOG = logging.getLogger("axolotl") @@ -60,7 +60,7 @@ def test_sharegpt_integration(self): ) as fin: data = fin.read() tokenized_conversation = json.loads(data) - prompter = ShareGPTPrompter("chat") + prompter = ShareGPTPrompterV2() strat = ShareGPTPromptTokenizingStrategy( prompter, self.tokenizer, @@ -79,7 +79,7 @@ def test_sharegpt_warnings_integration(self): ) as fin: data = fin.read() conversation = json.loads(data) - prompter = ShareGPTPrompter("chat") + prompter = ShareGPTPrompterV2() strat = ShareGPTPromptTokenizingStrategy( prompter, self.tokenizer, From 6d0268ee696506963b70b65e6f84b1aa12fbe361 Mon Sep 17 00:00:00 2001 From: Wing Lian Date: Mon, 25 Sep 2023 13:28:28 -0400 Subject: [PATCH 08/16] use sharegpt from prompt strategies now --- .../{sharegpt_simple.py => sharegpt.py} | 8 ++++++-- src/axolotl/utils/data.py | 11 ----------- 2 files changed, 6 insertions(+), 13 deletions(-) rename src/axolotl/prompt_strategies/{sharegpt_simple.py => sharegpt.py} (88%) diff --git a/src/axolotl/prompt_strategies/sharegpt_simple.py b/src/axolotl/prompt_strategies/sharegpt.py similarity index 88% rename from src/axolotl/prompt_strategies/sharegpt_simple.py rename to src/axolotl/prompt_strategies/sharegpt.py index a12ab1eab9..4647405d3d 100644 --- a/src/axolotl/prompt_strategies/sharegpt_simple.py +++ b/src/axolotl/prompt_strategies/sharegpt.py @@ -1,12 +1,16 @@ """Module containing the SimpleShareGPTPromptTokenizingStrategy class""" +from typing import Any, Dict, Optional from axolotl.prompt_tokenizers import ShareGPTPromptTokenizingStrategy from axolotl.prompters import ShareGPTPrompterV2 -def load(tokenizer, cfg): +def load(tokenizer, cfg, ds_cfg: Optional[Dict[str, Any]] = None): + conversation = ( + ds_cfg["conversation"] if ds_cfg and "conversation" in ds_cfg else None + ) return SimpleShareGPTPromptTokenizingStrategy( - ShareGPTPrompterV2(), + ShareGPTPrompterV2(conversation=conversation), tokenizer, cfg.train_on_inputs, cfg.sequence_len, diff --git a/src/axolotl/utils/data.py b/src/axolotl/utils/data.py index 461067b4ec..eeac15d302 100644 --- a/src/axolotl/utils/data.py +++ b/src/axolotl/utils/data.py @@ -25,7 +25,6 @@ GPTeacherPromptTokenizingStrategy, JeopardyPromptTokenizingStrategy, OpenAssistantPromptTokenizingStrategy, - ShareGPTPromptTokenizingStrategy, SummarizeTLDRPromptTokenizingStrategy, ) from axolotl.prompters import ( @@ -35,7 +34,6 @@ MultipleChoiceConcisePrompter, MultipleChoiceExplainPrompter, ReflectAlpacaPrompter, - ShareGPTPrompterV2, SummarizeTLDRPrompter, ) from axolotl.utils.dict import DictDefault @@ -320,15 +318,6 @@ def for_d_in_datasets(dataset_configs): ) ds_wrapper = TokenizedPromptDataset(ds_strategy, ds) datasets.append(ds_wrapper) - elif d_base_type == "sharegpt": - ds_strategy = ShareGPTPromptTokenizingStrategy( - ShareGPTPrompterV2(d.conversation), - tokenizer, - cfg.train_on_inputs, - cfg.sequence_len, - ) - ds_wrapper = TokenizedPromptDataset(ds_strategy, ds) - datasets.append(ds_wrapper) else: suffix = "" if ":load_" in d.type: From 7add5bb5bd1e3f2ab59d135a3f8f7ef897fb3a56 Mon Sep 17 00:00:00 2001 From: Wing Lian Date: Mon, 25 Sep 2023 13:51:41 -0400 Subject: [PATCH 09/16] update docs, add chatml template --- README.md | 7 ++++--- src/axolotl/prompt_strategies/sharegpt.py | 13 +++++++++++++ 2 files changed, 17 insertions(+), 3 deletions(-) diff --git a/README.md b/README.md index c2d4d8ef8c..3b3bc9985a 100644 --- a/README.md +++ b/README.md @@ -180,7 +180,7 @@ Have dataset(s) in one of the following format (JSONL recommended): ```json {"instruction": "...", "input": "...", "output": "..."} ``` -- `sharegpt:chat`: conversations where `from` is `human`/`gpt` +- `sharegpt`: conversations where `from` is `human`/`gpt` ```json {"conversations": [{"from": "...", "value": "..."}]} ``` @@ -269,11 +269,11 @@ Have dataset(s) in one of the following format (JSONL recommended): ```json {"prompt": "...", "generation": "..."} ``` -- `sharegpt_simple.load_role`: conversations where `role` is used instead of `from` +- `sharegpt.load_role`: conversations where `role` is used instead of `from` ```json {"conversations": [{"role": "...", "value": "..."}]} ``` -- `sharegpt_simple.load_guanaco`: conversations where `from` is `prompter`/`assistant` instead of default sharegpt +- `sharegpt.load_guanaco`: conversations where `from` is `prompter`/`assistant` instead of default sharegpt ```json {"conversations": [{"from": "...", "value": "..."}]} ``` @@ -443,6 +443,7 @@ datasets: data_files: # Optional[str] path to source data files shards: # Optional[int] number of shards to split data into name: # Optional[str] name of dataset configuration to load + conversation: # Optional[str] fastchat conversation type, only used with type: sharegpt # custom user prompt - path: repo diff --git a/src/axolotl/prompt_strategies/sharegpt.py b/src/axolotl/prompt_strategies/sharegpt.py index 4647405d3d..7dfb28e31e 100644 --- a/src/axolotl/prompt_strategies/sharegpt.py +++ b/src/axolotl/prompt_strategies/sharegpt.py @@ -1,9 +1,22 @@ """Module containing the SimpleShareGPTPromptTokenizingStrategy class""" from typing import Any, Dict, Optional +from fastchat.conversation import Conversation, SeparatorStyle, register_conv_template + from axolotl.prompt_tokenizers import ShareGPTPromptTokenizingStrategy from axolotl.prompters import ShareGPTPrompterV2 +register_conv_template( + Conversation( + name="chatml", + system_template="<|im_start|>system\n{system_message}", + system_message="You are a helpful assistant.", + roles=["<|im_start|>user", "<|im_start|>assistant"], + sep_style=SeparatorStyle.CHATML, + sep="<|im_end|>", + ) +) + def load(tokenizer, cfg, ds_cfg: Optional[Dict[str, Any]] = None): conversation = ( From 3989f84ee32d57325b720d7e677cd3113c183546 Mon Sep 17 00:00:00 2001 From: Wing Lian Date: Mon, 25 Sep 2023 14:01:02 -0400 Subject: [PATCH 10/16] add a newline after im_end token --- src/axolotl/prompt_strategies/sharegpt.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/src/axolotl/prompt_strategies/sharegpt.py b/src/axolotl/prompt_strategies/sharegpt.py index 7dfb28e31e..a0b8ba2731 100644 --- a/src/axolotl/prompt_strategies/sharegpt.py +++ b/src/axolotl/prompt_strategies/sharegpt.py @@ -13,7 +13,7 @@ system_message="You are a helpful assistant.", roles=["<|im_start|>user", "<|im_start|>assistant"], sep_style=SeparatorStyle.CHATML, - sep="<|im_end|>", + sep="<|im_end|>\n", ) ) From b541ffd3d0ea4656e649a7ed9d76f84f06d76de0 Mon Sep 17 00:00:00 2001 From: Wing Lian Date: Mon, 25 Sep 2023 14:18:08 -0400 Subject: [PATCH 11/16] ensure we correctly set system message --- src/axolotl/prompters.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/src/axolotl/prompters.py b/src/axolotl/prompters.py index 5b955cc0b9..3310f08389 100644 --- a/src/axolotl/prompters.py +++ b/src/axolotl/prompters.py @@ -250,7 +250,7 @@ def build_prompt(self, source) -> Generator[str, None, None]: # Add the conversation system prompt if provided, otherwise use the default one if source[0]["from"] == "system": - conv.system = source[0]["value"] + conv.set_system_message(source[0]["value"]) source.pop(0) roles = {"human": conv.roles[0], "gpt": conv.roles[1]} From 0cd7eb1abb8936e3fea3b0a6d46bb8c8dd6847cd Mon Sep 17 00:00:00 2001 From: Wing Lian Date: Tue, 26 Sep 2023 12:27:34 -0400 Subject: [PATCH 12/16] update per PR feedback to handle deprecated sharegpt types --- requirements.txt | 2 +- src/axolotl/utils/config.py | 19 +++++++++++++++++++ tests/test_validation.py | 23 +++++++++++++++++++++++ 3 files changed, 43 insertions(+), 1 deletion(-) diff --git a/requirements.txt b/requirements.txt index a3e87dacc9..3d7dc778ca 100644 --- a/requirements.txt +++ b/requirements.txt @@ -32,4 +32,4 @@ scikit-learn==1.2.2 pynvml art wandb -fschat +fschat==0.2.29 diff --git a/src/axolotl/utils/config.py b/src/axolotl/utils/config.py index 3a574cefcc..1dfdab2605 100644 --- a/src/axolotl/utils/config.py +++ b/src/axolotl/utils/config.py @@ -278,6 +278,25 @@ def validate_config(cfg): "`model_type: MixFormerSequentialForCausalLM` required for sample_packing" ) + if cfg.datasets: + for idx, ds_cfg in enumerate(cfg.datasets): + if ds_cfg.type == "sharegpt:chat": + LOG.warning( + PendingDeprecationWarning( + "`type: sharegpt:chat` will soon be deprecated. simply use `type: sharegpt` instead." + ) + ) + cfg.datasets[idx].type = "sharegpt" + if "sharegpt_simple" in ds_cfg.type: + LOG.warning( + PendingDeprecationWarning( + "`type: sharegpt_simple` will soon be deprecated. simply use `type: sharegpt` instead." + ) + ) + cfg.datasets[idx].type = cfg.datasets[idx].type.replace( + "sharegpt_simple", "sharegpt" + ) + # TODO # MPT 7b # https://github.com/facebookresearch/bitsandbytes/issues/25 diff --git a/tests/test_validation.py b/tests/test_validation.py index d7935c1a54..b9a57c2e9a 100644 --- a/tests/test_validation.py +++ b/tests/test_validation.py @@ -374,3 +374,26 @@ def test_merge_lora_no_bf16_fail(self): ) validate_config(cfg) + + def test_sharegpt_deprecation(self): + cfg = DictDefault( + {"datasets": [{"path": "lorem/ipsum", "type": "sharegpt:chat"}]} + ) + with self._caplog.at_level(logging.WARNING): + validate_config(cfg) + assert any( + "`type: sharegpt:chat` will soon be deprecated." in record.message + for record in self._caplog.records + ) + assert cfg.datasets[0].type == "sharegpt" + + cfg = DictDefault( + {"datasets": [{"path": "lorem/ipsum", "type": "sharegpt_simple:load_role"}]} + ) + with self._caplog.at_level(logging.WARNING): + validate_config(cfg) + assert any( + "`type: sharegpt_simple` will soon be deprecated." in record.message + for record in self._caplog.records + ) + assert cfg.datasets[0].type == "sharegpt:load_role" From 06ea94fed78110bbbe023c576a7a4345299433c7 Mon Sep 17 00:00:00 2001 From: Wing Lian Date: Tue, 26 Sep 2023 15:05:56 -0400 Subject: [PATCH 13/16] don't add duplicate wandb req --- requirements.txt | 1 - 1 file changed, 1 deletion(-) diff --git a/requirements.txt b/requirements.txt index 3d7dc778ca..7616d1fb0a 100644 --- a/requirements.txt +++ b/requirements.txt @@ -31,5 +31,4 @@ scipy scikit-learn==1.2.2 pynvml art -wandb fschat==0.2.29 From 52f9aef58ca803c7ad0b85855dc5c7b5160ac163 Mon Sep 17 00:00:00 2001 From: Wing Lian Date: Tue, 26 Sep 2023 15:16:11 -0400 Subject: [PATCH 14/16] make sharegpt fields configurable from yml --- src/axolotl/prompt_strategies/sharegpt.py | 8 +++++++- src/axolotl/prompters.py | 19 +++++++++++++++++-- 2 files changed, 24 insertions(+), 3 deletions(-) diff --git a/src/axolotl/prompt_strategies/sharegpt.py b/src/axolotl/prompt_strategies/sharegpt.py index a0b8ba2731..da36e778e0 100644 --- a/src/axolotl/prompt_strategies/sharegpt.py +++ b/src/axolotl/prompt_strategies/sharegpt.py @@ -22,8 +22,14 @@ def load(tokenizer, cfg, ds_cfg: Optional[Dict[str, Any]] = None): conversation = ( ds_cfg["conversation"] if ds_cfg and "conversation" in ds_cfg else None ) + field_human = ds_cfg["field_human"] if ds_cfg and "field_human" in ds_cfg else None + field_model = ds_cfg["field_model"] if ds_cfg and "field_model" in ds_cfg else None return SimpleShareGPTPromptTokenizingStrategy( - ShareGPTPrompterV2(conversation=conversation), + ShareGPTPrompterV2( + conversation=conversation, + role_key_model=field_model, + role_key_human=field_human, + ), tokenizer, cfg.train_on_inputs, cfg.sequence_len, diff --git a/src/axolotl/prompters.py b/src/axolotl/prompters.py index 3310f08389..61d0fa02c8 100644 --- a/src/axolotl/prompters.py +++ b/src/axolotl/prompters.py @@ -225,10 +225,15 @@ class ShareGPTPrompter: # pylint: disable=too-few-public-methods A prompter that generates prompts for the ShareGPT """ + role_key_human = "human" + role_key_model = "gpt" + def __init__( self, prompt_style=None, # pylint: disable=unused-argument conversation: Optional[Union[str, Conversation]] = None, + role_key_human: Optional[str] = None, + role_key_model: Optional[str] = None, ): if conversation: if isinstance(conversation, Conversation): @@ -237,6 +242,10 @@ def __init__( self._conversation = get_conv_template(conversation) else: self._conversation = get_conv_template("vicuna_v1.1") + if role_key_human: + self.role_key_human = role_key_human + if role_key_model: + self.role_key_model = role_key_model def build_prompt(self, source) -> Generator[str, None, None]: if len(source) < 2: @@ -253,7 +262,7 @@ def build_prompt(self, source) -> Generator[str, None, None]: conv.set_system_message(source[0]["value"]) source.pop(0) - roles = {"human": conv.roles[0], "gpt": conv.roles[1]} + roles = {self.role_key_human: conv.roles[0], self.role_key_model: conv.roles[1]} try: # Apply prompt templates @@ -287,5 +296,11 @@ class ShareGPTPrompterV2(ShareGPTPrompter): def __init__( self, conversation: Optional[Union[str, Conversation]] = None, + role_key_human: Optional[str] = None, + role_key_model: Optional[str] = None, ): - super().__init__(conversation=conversation) + super().__init__( + conversation=conversation, + role_key_human=role_key_human, + role_key_model=role_key_model, + ) From c312e9c25aac81a44fcf77d5ac70bf5b58d5bc6f Mon Sep 17 00:00:00 2001 From: Wing Lian Date: Tue, 26 Sep 2023 15:49:32 -0400 Subject: [PATCH 15/16] llama2 fixes --- src/axolotl/monkeypatch/fastchat_conversation_turns.py | 4 ++-- src/axolotl/prompters.py | 5 +---- 2 files changed, 3 insertions(+), 6 deletions(-) diff --git a/src/axolotl/monkeypatch/fastchat_conversation_turns.py b/src/axolotl/monkeypatch/fastchat_conversation_turns.py index 20a3356a27..19313fb7e2 100644 --- a/src/axolotl/monkeypatch/fastchat_conversation_turns.py +++ b/src/axolotl/monkeypatch/fastchat_conversation_turns.py @@ -85,9 +85,9 @@ def get_turns( # pylint: disable=too-many-return-statements if self.sep_style == SeparatorStyle.LLAMA2: seps = [self.sep, self.sep2] if self.system_message: - yield "", system_prompt + self.messages[0] + " " + yield "", system_prompt else: - yield "", "[INST] " + self.messages[0] + " " + yield "", "[INST] " for i, (role, message) in enumerate(self.messages[1:]): if message: yield role + " ", message + seps[i % 2] diff --git a/src/axolotl/prompters.py b/src/axolotl/prompters.py index 61d0fa02c8..ddccc793fa 100644 --- a/src/axolotl/prompters.py +++ b/src/axolotl/prompters.py @@ -266,10 +266,7 @@ def build_prompt(self, source) -> Generator[str, None, None]: try: # Apply prompt templates - if ( - source[0]["from"] not in roles - or roles[source[0]["from"]] != conv.roles[0] - ): + if source[0]["from"] not in roles: # Skip the first one if it is not from human source = source[1:] except IndexError as err: From 243d9e17f249d426b7e8c1535d6410634700908d Mon Sep 17 00:00:00 2001 From: Wing Lian Date: Tue, 26 Sep 2023 15:57:03 -0400 Subject: [PATCH 16/16] don't fail fatally when turns are improper --- src/axolotl/prompters.py | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/src/axolotl/prompters.py b/src/axolotl/prompters.py index ddccc793fa..370e15091c 100644 --- a/src/axolotl/prompters.py +++ b/src/axolotl/prompters.py @@ -276,7 +276,8 @@ def build_prompt(self, source) -> Generator[str, None, None]: conv.messages = [] for j, sentence in enumerate(source): role = roles[sentence["from"]] - assert role == conv.roles[j % 2], SHAREGPT_ASSERTION_FAILED_ROLE + if role != conv.roles[j % 2]: + LOG.warning(f"{SHAREGPT_ASSERTION_FAILED_ROLE}: {sentence}") conv.append_message(role, sentence["value"]) for part in conv.get_turns():