# What is wrong with my code

**URL:** </static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Ft%2Fwhat-is-wrong-with-my-code%2F113299%3E%5C
**Category:** Beginners\
**Created:** [October 22, 2024, 2:25am UTC](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Ft%2Fwhat-is-wrong-with-my-code%2F113299 "2024-10-22T02:25:44Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![Unrealaxolotl](https://avatars.discourse-cdn.com/v4/letter/u/46a35a/32.png) [@Unrealaxolotl](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Fu%2FUnrealaxolotl)%5C
**Post date:** [October 22, 2024, 2:25am UTC](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Ft%2Fwhat-is-wrong-with-my-code%2F113299%2F1 "2024-10-22T02:25:44Z")

</div>

import torch  
from datasets import Dataset  
from transformers import AdamW, AutoTokenizer, AutoModelForCausalLM, DataCollatorWithPadding,TrainingArguments, Trainer  
data = {  
“messages”: [“Hello, how are you?”, “What’s your name?”],  
“responses”: [“I’m fine, thank you!”, “I am a chatbot.”]  
}  
formatted\_data = [{“messages”: msg, “responses”: resp} for msg, resp in zip(data[“messages”], data[“responses”])]

raw\_dataset = Dataset.from\_list(formatted\_data)

checkpoint = “microsoft/DialoGPT-small”  
tokenizer = AutoTokenizer.from\_pretrained(checkpoint)  
tokenizer.pad\_token = tokenizer.eos\_token

def tokenize\_function(example):  
# Concatenate messages and responses  
inputs = tokenizer(example[“messages”], return\_tensors=“pt”, truncation=True,padding = True)  
# Create labels from responses (shifted input IDs)  
labels = tokenizer(example[“responses”], return\_tensors=“pt”, truncation=True,padding = True)[“input\_ids”]  
return {  
“input\_ids”: inputs[“input\_ids”][0], # Get the first batch item  
“attention\_mask”: inputs[“attention\_mask”][0], # Get the first batch item  
“labels”: labels[0] # Get the first batch item as labels  
}

tokenized\_datasets = raw\_dataset.map(tokenize\_function, batched=True)  
data\_collator = DataCollatorWithPadding(tokenizer=tokenizer)  
print(raw\_dataset)

#TRAINING TIME-----------------------------------

training\_args = TrainingArguments(“test-trainer”,report\_to=“none”)  
model = AutoModelForCausalLM.from\_pretrained(checkpoint)

from transformers import Trainer

trainer = Trainer(  
model,  
training\_args,  
train\_dataset=tokenized\_datasets,  
eval\_dataset=tokenized\_datasets,  
data\_collator=data\_collator,  
tokenizer=tokenizer,  
)

trainer.train()

model.save\_pretrained(“directory\_on\_my\_computer”)
