Hi HNs!

Co-founder of UbiAI here. We’ve just released UbiAI LLM Fine-tuning (https://ubiai.tools/fine-tuning), a platform to create high-quality training and fine-tune your LLM, all in one.

Starting out as a data labeling platform in 2021, we have witnessed firsthand the real challenge faced by many enterprises when trying to create high-quality datasets to train their AI model. From inconsistent labeling across annotators to unbalanced datasets and highly skewed labeling, we’ve seen it all. Manual data labeling is not only tedious but also time-consuming, very expensive and error-prone. For the last few years, we have worked closely with our customers to develop features that improve data quality and automate data labeling as much as possible with human-in-the loop. With LLMs becoming mainstream, accuracy, consistency and privacy are becoming real issues and more and more companies are leaning toward fine-tuning their own LLMs without real experience in creating high-quality data. In addition, the current fine-tuning platforms expect the users to bring their own dataset and don't offer a way to actually create the dataset which is the main bottleneck in the ML lifecycle.

To solve this issue, we have incorporated all of our learnings from the data labeling world into UbiAI LLM Fine-tuning to help companies to quickly create high quality data, fine-tune SOTA LLMs with few clicks and deploy it in production with a ready-to-use API endpoint.

The platform currently supports the following features: Intuitive data labeling to label any document (PDF, DOCX, TXT, CSV, etc.) Comprehensive collaboration feature to quickly create high-quality data Model finetuning: Currently we support Llama3.1, Mistral 7B, spaCy, BERT and LayoutLM Support multiple tasks: NER, Relation extraction, text classification, text completion (summarization, generation, text-to-SQL, etc.)

We have a free tier (https://ubiai.tools/pricing-ubiaiv2), so you can try it out for free. If you are looking to fine-tune your NLP or LLM model, we would love to hear from you!

If you have any questions or suggestions, please don’t hesitate to ask.