Oxen.ai makes fine-tuning easy, but if you are struggling to fine-tune your model or want us to fine-tune for you, weād be happy to set up a consultation with our ML experts!
1. Test Different Base Models


2. Upload or Create Your Dataset
Once you have found the right model, upload or create a dataset to fine-tune the model on. If you do not already have a dataset, you can explore new datasets and augment the data with our Model Inference tool. You can also use our Model Inference tool to generate synthetic data from scratch. If you already have a dataset, you can upload it easily with Oxen.aiās CLI commands.

3. Run The Base Model Through the Dataset
Before fine-tuning, it is crucial to evaluate your base model to see if the model is actually improving. We can do this with our Model Inference tool. First, go to your dataset, click āActionsā and select āRun Inferenceā.




4. Evaluate The Base Model
After we have the base model results, we now need to evaluate the quality of the base modelās responses to compare them with the fine-tuned responses. Using an LLM as a judge is a great option for a quick evaluation though nothing beats looking through the data yourself.We would use the Model Evals tool again. Go through the same process of opening the dataset and clicking the āModel Inferenceā button. This time, choose a different model, write a prompt explaining itās judging the quality of the responses, and pass in the prompt and response column. Weāre going to be using GPT-4o mini with the prompt:

Taking time to specify what you are looking for is important. Telling the model the exact criteria for what is good or bad will give you more accurate evaluations and control over the model accuracy. Itās also best practice to use a model from a different provider to evaluate the quality of the base modelās responses, since LLMs have been found to prefer their own responses even if the responses arenāt the best.
5. Fine-Tuning The Model
Now that we have our base model evaluated, go back to our training file, click āActionsā again, but this time click āFine-tune a modelā.





6. Next Steps
Now you have a finished fine-tuned model you can not only call via API, but you can chat directly to get a sense of how itās doing! Next, you can use Notebooks to run your model through your data, save the results, and compare it to the base model to see how itās improved. You can also:- Store your users questions and the model responses to create a ādata flywheelā and continuously improve your model.
- Keep fine-tuning different models to see which works best for your use case and data.
- Fine-tune on different datasets to see if another works better.
- Tweak your evaluation prompts to see if they are accurately evaluating the quality of your model.
Oxen.ai makes fine-tuning easy, but if you are struggling to fine-tune your model or want us to fine-tune for you, weād be happy to set up a consultation with our ML experts!