The world of large language models (LLMs) rapidly expanding, offering unprecedented opportunities for natural language processing across various domains. However, some critical challenges have remained unaddressed. At Giga ML, we proudly introduce the X1 Large 32k model, a pioneering on-premise LLM solution that addresses these critical issues.
Building on the llama2 70B model, we continued its pretraining and finetuned it using positional interpolation (as in Chen et al) to increase context length from 4k to 32k. The resulting X1 Large 32k model performs significantly better than the llama2 70B base model, and X1 Large performs close to Claude 2 on benchmarks like MT bench with a score of 8.4 vs 8.1. We are running detailed benchmarks. Our white paper will be out soon. Introducing this on-premise solution is a significant leap forward in the field of LLMs, and here's how:
1. Pretraining
Existing LLMs have been unable to pre-train on specific text data, a shortfall that significantly hampers their customization and effectiveness in specialized areas like healthcare, legal, and finance. With X1 Large, Giga ML empowers organizations to pre-train models with their domain-specific text data, enabling a much more tailored approach.
2. Finetuning
The ability to finetune LLMs for particular output structures or forms has been lacking, limiting their adaptability and application in areas that require specific and accurate responses. The X1 Large model offers the flexibility to finetune for your particular needs, ensuring more nuanced and relevant responses.
3. Privacy
Handling sensitive customer data with third-party servers like OpenAI and Anthropic has led to substantial security and privacy concerns. X1 Large overcomes these challenges by offering secure, on-premise solutions. Organizations can now handle sensitive data without compromising compliance or trust.
4. State-of-the-Art RAG with Mano AI: Unlocking New Opportunities
In partnership with Mano AI (YC S23), we are thrilled to unveil on-premise Retriever Augmented Generation (RAG). This novel collaboration paves the way for extensive, secure, and responsive document interaction. Mano AI takes care of On-premise embeddings and vector databases, so you don't need to worry about data privacy.
Coming Soon :
X1 - large legal
X1 - large Medical
X1 - large Finance
Last but not least, the most requested X1 large Code.
Read this blog if you want to know more about our enterprise offering.
Contact founders@gigaml.com to train and deploy the model on-premise in a week.