Unable to train safetensors format

Beginners
1

I’m trying to create a lora for the model from an example CSV of data, but we’re getting this error:

Traceback (most recent call last): File "/home/zino/apps/dload/main.py", line 49, in model = AutoModelForSequenceClassification.from_pretrained(model_name) File "/home/zino/apps/dload/.venv/lib/python3.10/site-packages/transformers/models/auto/auto_factory.py", line 484, in from_pretrained return model_class.from_pretrained( File "/home/zino/apps/dload/.venv/lib/python3.10/site-packages/transformers/modeling_utils.py", line 2555, in from_pretrained raise EnvironmentError( OSError: TheBloke/Nous-Hermes-13B-GPTQ does not appear to have a file named pytorch_model.bin, tf_model.h5, model.ckpt or flax_model.msgpack.

import pandas as pd import torch from torch.utils.data import Dataset, DataLoader from transformers import AutoTokenizer, AutoModelForSequenceClassification, AdamW from transformers import AutoModelForCausalLM import loralib as lora

Load CSV data

data = pd.read_csv(“data/guardian_articles.csv”)
train_data = data.sample(frac=0.8, random_state=1)
valid_data = data.drop(train_data.index)

Load tokenizer

model_name = “TheBloke/Nous-Hermes-13B-GPTQ”
#model_dir = “models/TheBloke_Nous-Hermes-13B-GPTQ”
tokenizer = AutoTokenizer.from_pretrained(model_name)

Tokenize data

def tokenize_data(data, tokenizer):
text_data = [str(text) for text in data[“bodyContent”].tolist()]
return tokenizer(text_data, padding=True, truncation=True, max_length=512)

train_encodings = tokenize_data(train_data, tokenizer)
valid_encodings = tokenize_data(valid_data, tokenizer)

Custom dataset class

class CustomDataset(Dataset):
def init(self, encodings, labels):
self.encodings = encodings
self.labels = labels

def __getitem__(self, idx):
    item = {key: torch.tensor(val[idx]) for key, val in self.encodings.items()}
    item["labels"] = torch.tensor(self.labels[idx])
    return item

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

train_dataset = CustomDataset(train_encodings, train_data[“webTitle”].tolist())
valid_dataset = CustomDataset(valid_encodings, valid_data[“webTitle”].tolist())

DataLoaders

train_loader = DataLoader(train_dataset, batch_size=8, shuffle=True)
valid_loader = DataLoader(valid_dataset, batch_size=8, shuffle=False)

Load pre-trained model

#model = AutoModelForCausalLM.from_pretrained(model_name)
model = AutoModelForSequenceClassification.from_pretrained(model_name)

Apply LoRA to the model

rank = 16
lora_model = lora.inject_lora(model, rank=rank)

Mark only LoRA parameters as trainable

lora.mark_only_lora_as_trainable(lora_model)

Optimizer

optimizer = AdamW(lora_model.parameters(), lr=1e-5)

Training loop

num_epochs = 3
device = torch.device(“cuda” if torch.cuda.is_available() else “cpu”)
lora_model.to(device)

for epoch in range(num_epochs):
lora_model.train()
for batch in train_loader:
optimizer.zero_grad()
input_ids = batch[“input_ids”].to(device)
attention_mask = batch[“attention_mask”].to(device)
labels = batch[“labels”].to(device)

    outputs = lora_model(input_ids, attention_mask=attention_mask, labels=labels)
    loss = outputs.loss
    loss.backward()
    optimizer.step()

Validation loop

lora_model.eval()
total_predictions = 0
correct_predictions = 0

with torch.no_grad():
for batch in valid_loader:
input_ids = batch[“input_ids”].to(device)
attention_mask = batch[“attention_mask”].to(device)
labels = batch[“labels”].to(device)

    outputs = lora_model(input_ids, attention_mask=attention_mask)
    predictions = torch.argmax(outputs.logits, dim=1)

    total_predictions += labels.size(0)
    correct_predictions += (predictions == labels).sum().item()

accuracy = correct_predictions / total_predictions
print(f"Validation accuracy: {accuracy:.2f}")

output_dir = “./lora_model”
lora_model.save_pretrained(output_dir)

Is it possible to use this format for training?