Finetune Llama-3.1-8B-Instruct
You can launch an fine-tune job with Llama-3.1-8B-Instruct using the quick-launch templates.
Finetuning templates come with pre-filled cost and performance optimized configurations. You can modify the config as per your requirements.
To launch the finetuning job, head over to the Fine-Tuning tab.
You can select Llama-3.1-8B from the available list of templates.
Alternatively, click on + Launch New Job button and choose Llama. Thereafter choose meta-llama/Llama-3.1-8B.
Enter your HuggingFace token. This will be used for pulling the model weights from the HF Llama repository.
All open-sourced Llama models by Meta require agreement to the community license agreement. If not done prior to launching the fine-tuning job, it will result in job failure.
Head over to the model card on HuggingFace and agree to the T&C to proceed.
For training dataset, you can choose a HuggingFace dataset or use your local/remote data store. For this tutorial, we'll use a HF dataset. You can checkout more about the dataset used here.
Click the Deploy Model button. The model will be up and running after a few minutes of provisioning.
And just like that, you have your own dedicated private Llama-3.1-8B-Instruct deployment. Once the deployment is in a Running state, you can take the model endpoint and model API key, plug it into the OpenAI SDK and query the model.