LLM Finetuning

With AutoTrain, you can easily finetune large language models (LLMs) on your own data!

AutoTrain supports the following types of LLM finetuning:

Causal Language Modeling (CLM)
Masked Language Modeling (MLM) [Coming Soon]

For LLM finetuning, only Hugging Face Hub model choice is available. User needs to select a model from Hugging Face Hub, that they want to finetune and select the parameters on their own (Manual Parameter Selection), or use AutoTrain’s Auto Parameter Selection to automatically select the best parameters for the task.

Data Preparation

LLM finetuning accepts data in CSV format. There are two modes for LLM finetuning: generic and chat. An example dataset with both formats in the same dataset can be found here: https://huggingface.co/datasets/tatsu-lab/alpaca

Generic

In generic mode, only one column is required: text. The user can take care of how the data is formatted for the task. A sample instance for this format is presented below:

Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request. 

### Instruction: Evaluate this sentence for spelling and grammar mistakes 

### Input: He finnished his meal and left the resturant 

### Response: He finished his meal and left the restaurant.

Generic LLM Finetuning

Please note that above is the format for instruction finetuning. But in the generic mode, you can also finetune on any other format as you want. The data can be changed according to the requirements.

Training

Once you have your data ready and estimate verified, you can start training your model by clicking the “Create Project” button.