Blenderbot: Schlüsselfehler beim Training von JSON ⇐ Python

[Anonymous]

Ich versuche, das Blenderbot-Modell facebook/blenderbot-90m auf einer Reihe von Fragebögen basierend auf meinem Eingabe-JSON zu trainieren.
Das Beispiel-JSON (Snippet) sieht so aus dies:

Code: Select all

[
{
"input": "Who is Carole Nicholas?",
"output": "She is a software engineer !"
},
{
"input": "Where was Carole Nicholas born?",
"output": "She was born in San Franscisco."
},
{
"input": "Which was Carole's school?",
"output": "Lake Side school."
},
{
"input": "Where is Lakeside school located?",
"output": "It is located at Seattle!"
}
]

Code für die Schulung:

Code: Select all

from transformers import AutoTokenizer, AutoModelForSeq2SeqLM, DataCollatorWithPadding
from datasets import Dataset
import torch

modelName = "facebook/blenderbot-90m"

# Load the dataset
dataset = Dataset.from_json("C:\\Users\\Raj\\Desktop\\To-Dos\\AI_related\\chatInput.json")

model = AutoModelForSeq2SeqLM.from_pretrained(modelName)
tokenizer = AutoTokenizer.from_pretrained(modelName)

# Preprocess the data
def preprocess(examples):
model_inputs = tokenizer(
examples["input"],
examples["output"],
padding="max_length",
truncation=True,
max_length=512,
)
return model_inputs

dataset = dataset.map(preprocess, batched=True)
#print(dataset['input'])

train_size = int(len(dataset) * 0.8)
val_size = int(len(dataset) * 0.1)
test_size = len(dataset) - train_size - val_size

data_collator = DataCollatorWithPadding(tokenizer=tokenizer)

from torch.utils.data import Dataset

class BlenderBotDataset(Dataset):
def __init__(self, dataset, tokenizer):
self.dataset = dataset
self.tokenizer = tokenizer

def __len__(self):
return len(self.dataset)

def __getitem__(self, idx):
encoding = self.dataset[idx]
return {
"input_ids": torch.tensor(encoding["input_ids"]),
"attention_mask": torch.tensor(encoding["attention_mask"]),
"labels": torch.tensor(encoding["input_ids"]),
}

from torch.utils.data import DataLoader

train_dataset = BlenderBotDataset(dataset[train_size], tokenizer)
val_dataset = BlenderBotDataset(dataset[val_size], tokenizer)

train_loader = DataLoader(train_dataset, batch_size=16, shuffle=True)
val_loader = DataLoader(val_dataset, batch_size=16, shuffle=False)

from transformers import Trainer, TrainingArguments

# Define the training arguments
#training_args = TrainingArguments("test-trainer")

training_args = TrainingArguments(
output_dir="C:\\Users\\Raj\\Desktop\\To-Dos\\AI_related",
num_train_epochs=3,
per_device_train_batch_size=16,
per_device_eval_batch_size=64,
evaluation_strategy="epoch",
learning_rate=5e-5,
save_total_limit=2,
save_steps=500,
#load_best_model_at_end=True,
metric_for_best_model="accuracy",
greater_is_better=True,
save_on_each_node=True
)

print(dataset[train_size])

# Create the trainer
trainer = Trainer(
model,
args=training_args,
train_dataset=train_dataset,
eval_dataset=val_dataset,
#tokenizer = tokenizer,
data_collator=data_collator
#compute_metrics=lambda pred: {"accuracy":  torch.sum(torch.argmax(pred.label_ids, dim=1) == torch.argmax(pred.predictions, dim=1)) / len(pred.label_ids)},
)

# Train the model
trainer.train()

Gibt diesen Fehler aus:

Code: Select all

0%|          | 0/3 [00:00