Finefuning LLaMA2 model using autotrain advanced

Beginners
1

Hello, I am trying to finetune llama2 7B model with my own dataset.
I am stuck at setting the data-path correctly. Here are my attempts:

1st command:
autotrain llm --train --data-path ./data --text-column text --peft --auto_find_batch_size --epochs 3 --trainer sft --model meta-llama/Llama-2-7b-hf --project-name ftllama2

=> Somehow, the data_path got changed to “ftllama2/autotrain-data”

2nd command:
autotrain llm --train --data-path ./data/train.csv --text-column text --peft --auto_find_batch_size --epochs 3 --trainer sft --model meta-llama/Llama-2-7b-hf --project-name ftllama2

=> The data_path is correct, but I am getting this error:
:x: ERROR | 2024-02-15 08:44:06 | autotrain.trainers.common:wrapper:92 - Couldn’t find a dataset script at /home/ubuntu/workspace/git/language-agnostic-embedding/data/train.csv/train.csv.py or any data file in the same directory.

Any comments are welcomed.

2

Hello, I also got this error before, you need to upload the dataset in your huggingface account and then call the created dataset from huggingface in the data-path parameter. For an instance, it will be like this : your-user-name/the-name-of-your-created-dataset.
I hope this helps!