Alphatross LLM Now Available on Hugging Face
Jan 26, 2024
Designed for Financial Institutions
Since introducing Albatross in October, we’ve seen the profound impact that it’s had on financial institutions that are looking to accelerate AI transformation. Today we’ve made
v-alpha-tross, an earlier version of Gradient's Albatross model with limited capabilities, available on Hugging Face.
v-alpha-trossavailable openly, we believe this can benefit everyone in the community. Giving businesses, startups, entrepreneurs, and researchers access to a model that would be challenging to build themselves and compute power that they might not otherwise have access to. This accessibility paves the way for groundbreaking experimentation and innovation within financial services, offering both economic and social advantages.
v-alpha-tross model is an early version of our Albatross model with limited capabilities, specializing in financial services. Built on Meta’s Llama 2 foundational model,
v-alpha-tross has been extensively optimized with finance specific pre-training, fine-tuning and instruction tuning. As a result,
v-alpha-trossovercomes deficiencies that general-purpose language models face when solving domain-specific tasks in finance.
v-alpha-tross is currently the highest performing H6 model for Llama2-70B variations and substantially outperforms similar variations on GSM8K. Designed specifically for financial services,
v-alpha-tross reaches perfect performance in financial tasks like extracting information from tabular data found in SEC filings.
Furthering Innovation in Financial Services
At Gradient, we are excited to give back to the community and inspire new and innovative solutions within financial services. We believe that by openly sharing our domain-specific model in finance, it will support the development of helpful and safer generative AI. As we continue our mission to democratize AI, we look forward to seeing what the world builds with
As always, Gradient will continue to put our best foot forward as we navigate responsible AI - focusing on the following four pillars: logic & reasoning, privacy, security, and regulation & compliance.