# 必要なものをインポートします
import torch
import math
import random
import logging
import warnings
from collections import defaultdict
from transformers import T5Tokenizer, GPT2LMHeadModel, GPT2Config, AutoModelForCausalLM, GPT2DoubleHeadsModel
from transformers.optimization import AdamW, get_linear_schedule_with_warmup
import pandas as pd
from itertools import chain
import os
from tqdm.auto import tqdm, trange
from simpletransformers.config.model_args import ConvAIArgs
from simpletransformers.conv_ai.conv_ai_utils import get_dataset
from simpletransformers.config.utils import sweep_config_to_sweep_values
import wandb
import torch.nn.functional as F
from torch.utils.tensorboard import SummaryWriter
from torch.utils.data import DataLoader, RandomSampler, SequentialSampler, TensorDataset
from sklearn.metrics import (
f1_score
)
logger = logging.getLogger(__name__)
SPECIAL_TOKENS = ["<s>", "</s>", "<speaker1>", "<speaker2>", "[PAD]"]
ATTR_TO_SPECIAL_TOKEN = {
"bos_token": "<s>",
"eos_token": "</s>",
"pad_token": "<PAD>",
"additional_special_tokens": ["<speaker1>", "<speaker2>"],
}
MODEL_INPUTS = ["input_ids", "mc_token_ids", "labels", "mc_labels", "token_type_ids"]
PADDED_INPUTS = ["input_ids", "labels", "token_type_ids"]
class ConvAIModel:
def __init__(self, model_name, args=None, **kwargs):
self.args = self._load_model_args(model_name)
if "sweep_config" in kwargs:
self.is_sweeping = True
sweep_config = kwargs.pop("sweep_config")
sweep_values = sweep_config_to_sweep_values(sweep_config)
self.args.update_from_dict(sweep_values)
else:
self.is_sweeping = False
if isinstance(args, dict):
self.args.update_from_dict(args)
elif isinstance(args, ConvAIArgs):
self.args = args
self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
# 日本語版のモデルに合わせる
self.tokenizer = T5Tokenizer.from_pretrained(model_name)
self.tokenizer.do_lower_case = True
# AutoModelForCausalLM
self.model = GPT2DoubleHeadsModel.from_pretrained(model_name)
self.config = GPT2Config.from_pretrained(model_name)
self.add_special_tokens_(self.model, self.tokenizer)
def train_model(self, train_file=None, output_dir=None, show_running_loss=True, args=None, eval_file=None, verbose=True, **kwargs):
if self.args.evaluate_during_training and eval_file is None:
warnings.warn(
"eval_file not specified but evaluate_during_training is True. Using personachat eval data."
)
if args:
self.args.update_from_dict(args)
if not output_dir:
output_dir = self.args.output_dir
if (
os.path.exists(output_dir)
and os.listdir(output_dir)
and not self.args.overwrite_output_dir
):
raise ValueError(
"Output directory ({}) already exists and is not empty."
" Set overwrite_output_dir: True to automatically overwrite.".format(
output_dir
)
)
self._move_model_to_device()
train_dataloader, train_sampler = self.load_and_cache_examples(
dataset_path=train_file,
verbose=verbose,
no_cache=self.args.no_cache or self.args.reprocess_input_data,
)
eval_loader = None
os.makedirs(output_dir, exist_ok=True)
global_step, training_details = self.train(
train_dataloader,
output_dir,
show_running_loss=show_running_loss,
verbose=verbose,
**kwargs,
)
self.save_model(model=self.model)
if verbose:
logger.info(
" Training of {} model complete. Saved to {}.".format(
self.args.model_type, output_dir
)
)
def train(self, train_dataloader, output_dir, show_running_loss=True, verbose=True, **kwargs,):
device = self.device
model = self.model
args = self.args
tb_writer = SummaryWriter(log_dir=args.tensorboard_dir)
t_total = (
len(train_dataloader)
// args.gradient_accumulation_steps
* args.num_train_epochs
)
no_decay = ["bias", "LayerNorm.weight"]
optimizer_grouped_parameters = []
custom_parameter_names = set()
for group in self.args.custom_parameter_groups:
params = group.pop("params")
custom_parameter_names.update(params)
param_group = {**group}
param_group["params"] = [
p for n, p in model.named_parameters() if n in params
]
optimizer_grouped_parameters.append(param_group)
for group in self.args.custom_layer_parameters:
layer_number = group.pop("layer")
layer = f"layer.{layer_number}."
group_d = {**group}
group_nd = {**group}
group_nd["weight_decay"] = 0.0
params_d = []
params_nd = []
for n, p in model.named_parameters():
if n not in custom_parameter_names and layer in n:
if any(nd in n for nd in no_decay):
params_nd.append(p)
else:
params_d.append(p)
custom_parameter_names.add(n)
group_d["params"] = params_d
group_nd["params"] = params_nd
optimizer_grouped_parameters.append(group_d)
optimizer_grouped_parameters.append(group_nd)
if not self.args.train_custom_parameters_only:
optimizer_grouped_parameters.extend(
[
{
"params": [
p
for n, p in model.named_parameters()
if n not in custom_parameter_names
and not any(nd in n for nd in no_decay)
],
"weight_decay": args.weight_decay,
},
{
"params": [
p
for n, p in model.named_parameters()
if n not in custom_parameter_names
and any(nd in n for nd in no_decay)
],
"weight_decay": 0.0,
},
]
)
warmup_steps = math.ceil(t_total * args.warmup_ratio)
args.warmup_steps = (
warmup_steps if args.warmup_steps == 0 else args.warmup_steps
)
optimizer = AdamW(
optimizer_grouped_parameters,
lr=args.learning_rate,
eps=args.adam_epsilon,
)
scheduler = get_linear_schedule_with_warmup(
optimizer,
num_warmup_steps=args.warmup_steps,
num_training_steps=t_total,
)
if args.n_gpu > 1:
model = torch.nn.DataParallel(model)
global_step = 0
training_progress_scores = None
tr_loss, logging_loss = 0.0, 0.0
model.zero_grad()
train_iterator = trange(
int(args.num_train_epochs), desc="Epoch", disable=args.silent
)
epoch_number = 0
best_eval_metric = None
early_stopping_counter = 0
if args.fp16:
from torch.cuda import amp
scaler = amp.GradScaler()
for _ in train_iterator:
model.train()
train_iterator.set_description(
f"Epoch {epoch_number} of {args.num_train_epochs}"
)
batch_iterator = tqdm(
train_dataloader,
desc=f"Running Epoch {epoch_number + 1} of {args.num_train_epochs}",
disable=args.silent,
mininterval=0,
)
for step, batch in enumerate(batch_iterator):
batch = tuple(t.to(device) for t in batch)
input_ids, mc_token_ids, labels, mc_labels, token_type_ids = batch
if args.fp16:
with amp.autocast():
outputs = model(
input_ids,
token_type_ids=token_type_ids,
mc_token_ids=mc_token_ids,
mc_labels=mc_labels,
labels=labels,
)
lm_loss, mc_loss = outputs[:2]
# model outputs are always tuple in pytorch-transformers (see doc)
loss = lm_loss * args.lm_coef + mc_loss * args.mc_coef
else:
outputs = model(
input_ids,
token_type_ids=token_type_ids,
mc_token_ids=mc_token_ids,
mc_labels=mc_labels,
labels=labels,
)
lm_loss, mc_loss = outputs[:2]
# model outputs are always tuple in pytorch-transformers (see doc)
loss = lm_loss * args.lm_coef + mc_loss * args.mc_coef
if args.n_gpu > 1:
loss = (
loss.mean()
)
current_loss = loss.item()
if show_running_loss:
print("\rRunning loss: %f" % current_loss, end="")
if args.gradient_accumulation_steps > 1:
loss = loss / args.gradient_accumulation_steps
if args.fp16:
scaler.scale(loss).backward()
else:
loss.backward()
tr_loss += loss.item()
if (step + 1) % args.gradient_accumulation_steps == 0:
if args.fp16:
scaler.unscale_(optimizer)
if args.optimizer == "AdamW":
torch.nn.utils.clip_grad_norm_(
model.parameters(), args.max_grad_norm
)
if args.fp16:
scaler.step(optimizer)
scaler.update()
else:
optimizer.step()
scheduler.step() # Update learning rate schedule
model.zero_grad()
global_step += 1
if args.logging_steps > 0 and global_step % args.logging_steps == 0:
# Log metrics
tb_writer.add_scalar(
"lr", scheduler.get_last_lr()[0], global_step
)
tb_writer.add_scalar(
"loss",
(tr_loss - logging_loss) / args.logging_steps,
global_step,
)
logging_loss = tr_loss
if args.wandb_project or self.is_sweeping:
wandb.log(
{
"Training loss": current_loss,
"lr": scheduler.get_last_lr()[0],
"global_step": global_step,
}
)
if args.save_steps > 0 and global_step % args.save_steps == 0:
# Save model checkpoint
output_dir_current = os.path.join(
output_dir, "checkpoint-{}".format(global_step)
)
self.save_model(output_dir_current, model=model)
epoch_number += 1
output_dir_current = os.path.join(
output_dir, "checkpoint-{}-epoch-{}".format(global_step, epoch_number)
)
if args.save_model_every_epoch or args.evaluate_during_training:
os.makedirs(output_dir_current, exist_ok=True)
if args.save_model_every_epoch:
self.save_model(output_dir_current, model=model)
return (
global_step,
tr_loss / global_step
if not self.args.evaluate_during_training
else training_progress_scores,
)
def load_and_cache_examples(self, dataset_path=None, evaluate=False, no_cache=False, verbose=True, silent=False):
process_count = self.args.process_count
tokenizer = self.tokenizer
args = self.args
if not no_cache:
no_cache = args.no_cache
os.makedirs(self.args.cache_dir, exist_ok=True)
dataset_path = dataset_path if dataset_path else ""
dataset = get_dataset(
tokenizer,
dataset_path,
args.cache_dir,
process_count=process_count,
proxies=self.__dict__.get("proxies", None),
evaluate=evaluate,
no_cache=no_cache,
args=args,
)
datasets = defaultdict(list)
num_candidates = len(dataset[0]["utterances"][0]["candidates"])
if args.num_candidates > 0 and not evaluate:
num_candidates = min(args.num_candidates, num_candidates)
for dialog in dataset:
persona = dialog["personality"].copy()
for _ in range(args.personality_permutations):
for utterance in dialog["utterances"]:
history = utterance["history"][-(2 * args.max_history + 1) :]
for j, candidate in enumerate(
utterance["candidates"][-num_candidates:]
):
labels = bool(j == num_candidates - 1)
instance = self.build_input_from_segments(
persona, history, candidate, tokenizer, labels
)
for input_name, input_array in instance.items():
datasets[input_name].append(input_array)
datasets["mc_labels"].append(num_candidates - 1)
datasets["n_candidates"] = num_candidates
persona = [persona[-1]] + persona[:-1] # permuted personalities
tensor_datasets = []
dataset = self.pad_dataset(
datasets, padding=tokenizer.convert_tokens_to_ids(SPECIAL_TOKENS[-1])
)
for input_name in MODEL_INPUTS:
tensor = torch.tensor(dataset[input_name])
if input_name != "mc_labels":
tensor = tensor.view((-1, datasets["n_candidates"]) + tensor.shape[1:])
tensor_datasets.append(tensor)
tensor_dataset = TensorDataset(*tensor_datasets)
if not evaluate:
data_sampler = RandomSampler(tensor_dataset)
data_loader = DataLoader(
tensor_dataset, sampler=data_sampler, batch_size=args.train_batch_size
)
else:
data_sampler = SequentialSampler(tensor_dataset)
data_loader = DataLoader(
tensor_dataset, sampler=data_sampler, batch_size=args.eval_batch_size
)
return data_loader, data_sampler
def compute_metrics(self, mc_preds, mc_labels, lm_logits, labels, **kwargs):
loss_fct = torch.nn.CrossEntropyLoss(ignore_index=-100)
extra_metrics = {}
for metric, func in kwargs.items():
extra_metrics[metric] = func(mc_labels, mc_preds)
f1_current = f1_score(mc_labels.cpu().numpy(), mc_preds, average="macro")
lm_loss_current = loss_fct(lm_logits, labels)
return {
**{"f1_score": f1_current, "language_model_loss": lm_loss_current},
**extra_metrics,
}
def interact(self, personality=None):
model = self.model
args = self.args
tokenizer = self.tokenizer
process_count = self.args.process_count
if self.args.fp16:
from torch.cuda import amp
self._move_model_to_device()
if not personality:
dataset = get_dataset(
tokenizer,
None,
args.cache_dir,
process_count=process_count,
proxies=self.__dict__.get("proxies", None),
interact=True,
args=args,
)
personalities = [
dialog["personality"]
for dataset in dataset.values()
for dialog in dataset
]
personality = random.choice(personalities)
else:
personality = [tokenizer.encode(s.lower()) for s in personality]
history = []
while True:
raw_text = input(">>> ")
while not raw_text:
print("Prompt should not be empty!")
raw_text = input(">>> ")
history.append(
tokenizer.encode(raw_text)
)
with torch.no_grad():
if args.fp16:
with amp.autocast():
out_ids = self.sample_sequence(
personality, history, tokenizer, model, args
)
else:
out_ids = self.sample_sequence(
personality, history, tokenizer, model, args
)
history.append(out_ids)
history = history[-(2 * args.max_history + 1) :]
out_text = tokenizer.decode(
out_ids, skip_special_tokens=self.args.skip_special_tokens
)
print("you->", raw_text)
print("bot->", out_text)
print("--------------------------------")
# print(history)
def interact_single(self, message, history, personality=None, encode_history=True):
model = self.model
args = self.args
tokenizer = self.tokenizer
process_count = self.args.process_count
if self.args.fp16:
from torch.cuda import amp
self._move_model_to_device()
if not personality:
dataset = get_dataset(
tokenizer,
None,
args.cache_dir,
process_count=process_count,
proxies=self.__dict__.get("proxies", None),
interact=True,
)
personalities = [
dialog["personality"]
for dataset in dataset.values()
for dialog in dataset
]
personality = random.choice(personalities)
else:
personality = [tokenizer.encode(s.lower()) for s in personality]
if encode_history:
raw_history = history.copy()
raw_history.append(message)
history = [tokenizer.encode(sentence) for sentence in history]
history.append(tokenizer.encode(message))
with torch.no_grad():
if args.fp16:
with amp.autocast():
out_ids = self.sample_sequence(
personality, history, tokenizer, model, args
)
else:
out_ids = self.sample_sequence(
personality, history, tokenizer, model, args
)
out_text = tokenizer.decode(
out_ids, skip_special_tokens=self.args.skip_special_tokens
)
if encode_history:
raw_history.append(out_text)
history = raw_history
else:
history.append(out_ids)
return out_text, history
def _threshold(self, x, threshold):
if x >= threshold:
return 1
return 0
def _move_model_to_device(self):
self.model.to(self.device)
def _get_last_metrics(self, metric_values):
return {metric: values[-1] for metric, values in metric_values.items()}
def _create_training_progress_scores(self, **kwargs):
extra_metrics = {key: [] for key in kwargs}
training_progress_scores = {
"global_step": [],
"language_model_loss": [],
"f1_score": [],
**extra_metrics,
}
return training_progress_scores
def save_model(self, output_dir=None, model=None, results=None):
if not output_dir:
output_dir = self.args.output_dir
if model and not self.args.no_save:
# Take care of distributed/parallel training
model_to_save = model.module if hasattr(model, "module") else model
model_to_save.save_pretrained(output_dir)
self.tokenizer.save_pretrained(output_dir)
self.save_model_args(output_dir)
if results:
output_eval_file = os.path.join(output_dir, "eval_results.txt")
with open(output_eval_file, "w") as writer:
for key in sorted(results.keys()):
writer.write("{} = {}\n".format(key, str(results[key])))
def add_special_tokens_(self, model, tokenizer):
orig_num_tokens = 32000
num_added_tokens = tokenizer.add_special_tokens(
ATTR_TO_SPECIAL_TOKEN
) # doesn't add if they are already there
if num_added_tokens > 0:
self.model.resize_token_embeddings(
new_num_tokens=orig_num_tokens + num_added_tokens
)
def build_input_from_segments(self, persona, history, reply, tokenizer, labels=False, with_eos=True):
bos, eos, speaker1, speaker2 = tokenizer.convert_tokens_to_ids(
SPECIAL_TOKENS[:-1]
)
sequence = (
[[bos] + list(chain(*persona))]
+ history
+ [reply + ([eos] if with_eos else [])]
)
sequence = [sequence[0]] + [
[speaker2 if (len(sequence) - i) % 2 else speaker1] + s
for i, s in enumerate(sequence[1:])
]
instance = {}
instance["input_ids"] = list(chain(*sequence))
instance["token_type_ids"] = [
speaker2 if i % 2 else speaker1 for i, s in enumerate(sequence) for _ in s
]
instance["mc_token_ids"] = len(instance["input_ids"]) - 1
instance["labels"] = [-100] * len(instance["input_ids"])
if labels:
instance["labels"] = (
([-100] * sum(len(s) for s in sequence[:-1]))
+ [-100]
+ sequence[-1][1:]
)
return instance
def pad_dataset(self, dataset, padding=0):
max_l = max(len(x) for x in dataset["input_ids"])
for name in PADDED_INPUTS:
dataset[name] = [
x + [padding if name != "labels" else -100] * (max_l - len(x))
for x in dataset[name]
]
return dataset
def top_filtering(self, logits, top_k=0.0, top_p=0.9, threshold=-float("Inf"), filter_value=-float("Inf"), ):
assert (
logits.dim() == 1
) # Only work for batch size 1 for now - could update but it would obfuscate a bit the code
top_k = min(top_k, logits.size(-1))
if top_k > 0:
# Remove all tokens with a probability less than the last token in the top-k tokens
indices_to_remove = logits < torch.topk(logits, top_k)[0][..., -1, None]
logits[indices_to_remove] = filter_value
if top_p > 0.0:
# Compute cumulative probabilities of sorted tokens
sorted_logits, sorted_indices = torch.sort(logits, descending=True)
cumulative_probabilities = torch.cumsum(
F.softmax(sorted_logits, dim=-1), dim=-1
)
# Remove tokens with cumulative probability above the threshold
sorted_indices_to_remove = cumulative_probabilities > top_p
# Shift the indices to the right to keep also the first token above the threshold
sorted_indices_to_remove[..., 1:] = sorted_indices_to_remove[
..., :-1
].clone()
sorted_indices_to_remove[..., 0] = 0
# Back to unsorted indices and set them to -infinity
indices_to_remove = sorted_indices[sorted_indices_to_remove]
logits[indices_to_remove] = filter_value
indices_to_remove = logits < threshold
logits[indices_to_remove] = filter_value
return logits
def sample_sequence(self, personality, history, tokenizer, model, args, current_output=None):
special_tokens_ids = tokenizer.convert_tokens_to_ids(SPECIAL_TOKENS)
if current_output is None:
current_output = []
for i in range(args.max_length):
instance = self.build_input_from_segments(
personality, history, current_output, tokenizer, with_eos=False
)
input_ids = torch.tensor(
instance["input_ids"], device=self.device
).unsqueeze(0)
token_type_ids = torch.tensor(
instance["token_type_ids"], device=self.device
).unsqueeze(0)
logits = model(input_ids, token_type_ids=token_type_ids)
logits = logits[0]
logits = logits[0, -1, :] / args.temperature
logits = self.top_filtering(logits, top_k=args.top_k, top_p=args.top_p)
probs = F.softmax(logits, dim=-1)
prev = (
torch.topk(probs, 1)[1]
if not args.do_sample
else torch.multinomial(probs, 1)
)
if i < args.min_length and prev.item() in special_tokens_ids:
while prev.item() in special_tokens_ids:
if probs.max().item() == 1:
warnings.warn(
"Warning: model generating special token with probability 1."
)
break # avoid infinitely looping over special token
prev = torch.multinomial(probs, num_samples=1)
if prev.item() in special_tokens_ids:
break
current_output.append(prev.item())
return current_output
def save_model_args(self, output_dir):
os.makedirs(output_dir, exist_ok=True)
self.args.save(output_dir)
def _load_model_args(self, input_dir):
args = ConvAIArgs()
args.load(input_dir)
return args
def get_named_parameters(self):
return [n for n, p in self.model.named_parameters()]
コメント
コメント一覧 （4件）
非常に面白い記事でした！
自分の環境で再現するにあたって質問したいことがあります。
記事で紹介されていた使用例を自分の環境で再現しようとしたのですが、
同じ返答が連続するなど変な挙動を見せました。
“`
you-> 私の名前はhogeです。
bot-> 私はバトーさんの魔法が好きだよ。 僕はバトーさんの魔法が好きだよ
——————————–
>>> あなたの名前は何ですか？
you-> あなたの名前は何ですか？
bot-> 僕はバトーさんの魔法が好きだよ。 僕はバトーさんの魔法が好き
——————————–
>>> 魔法って何？
you-> 魔法って何？
bot-> 僕はバトーさんの魔法が好きだ。 僕はバトーさんの魔法が好き
“`
zanさんはこういった現象に見舞われたことはありますか？
質問ありがとうございます。
私の場合も質問によっては同じような返答が連続することがありました。
データ量を増やしたり、少し工夫（話題が広がるような会話データを作るなど）したりすると、ある程度はこのような挙動が減少しました。
色々試してみてください！
すみません。
こちらの記事のコードを実行したいのですが、いろいろ分からず四苦八苦しています。
まず、model codeをgoole colaboratoryで実行した際に、
no module named transformersと表示され、それよりも先に進めずにいます。
また、途中からrun.ipynbでコードを実行していますが、当初googleドライブをマウントしたipynbからどう実行していますか？
できればこのあたりのやり方について、詳しく教えていただけるとありがたいです。
コメントありがとうございます。
no module named transformersについては、transformersというモジュールがないよというエラーなので、!pip install simpletransformersを実行してsimpletransformersをインストールすることで一緒にインストールされて解決すると思います。実行する場所については、追加しているのでそちらを参考にしてください。途中からrun.ipynbにコードをコピペして実行するのではなく、すべてのコードをrun.ipynbにコピペして順番に実行して見てください。どのファイルで「フォルダを作成」のコードを実行するのかについて追加しています。
まだ何かエラーが出るようでしたらコメントをお願いします。