Abacus.AI Releases Open-Source AI Model ‘Dracarys’ for Coding

Generative AI for coding gets a boost: Abacus.ai has released Dracarys, a new family of open-source large language models (LLMs) specifically optimized for coding tasks.

• Dracarys applies a fine-tuning “recipe” to existing LLMs, significantly improving their coding abilities.
• The initial release focuses on the 70B parameter class of models, including Meta’s Llama 3.1 and Qwen-2 72B.
• Abacus.ai has previously released Smaug-72B, a general-purpose LLM, showcasing their penchant for using dragon-inspired names from popular fiction.

Performance improvements: Benchmarks indicate substantial enhancements in coding capabilities for models treated with the Dracarys recipe.

• LiveBench coding scores show an increase from 32.67 to 35.23 for the meta-llama-3.1-70b-instruct turbo model after applying the Dracarys recipe.
• The qwen2-72b-instruct model saw an even more significant boost, with its coding score rising from 32.38 to 38.95.
• Bindu Reddy, CEO and co-founder of Abacus.ai, claims that Dracarys-72B-Instruct is now the best coding model in its class among open-source options.

Competitive landscape: The release of Dracarys enters a growing market for generative AI in application development and coding.

• GitHub Copilot pioneered the space, offering code completion and development assistance.
• Startups like Tabnine and Replit have been developing similar LLM-powered features for developers.
• Major LLM providers are also competing in this space, with Anthropic’s Claude 3.5 Sonnet emerging as a popular choice for coding tasks in 2024.

Availability and future plans: Abacus.ai is making Dracarys accessible through various channels and plans to expand its application.

• Model weights for both Llama and Qwen2-based Dracarys versions are available on Hugging Face.
• The fine-tuned models are part of Abacus.ai’s Enterprise offering, providing an option for companies concerned about data privacy when using public APIs.
• Abacus.ai is considering making Dracarys available on their ChatLLM service for small teams and professionals if there’s sufficient demand.
• Future plans include releasing Dracarys versions for Deepseek-coder and Llama-3.1 400b models.

Implications for enterprise coding: The improved performance of Dracarys models could have significant benefits for developers and enterprises.

• Enhanced coding capabilities may lead to increased productivity and efficiency in software development processes.
• The availability of open-source models with improved coding abilities provides alternatives to closed-source solutions, potentially reducing dependency on specific vendors.
• Enterprises concerned about data privacy when using public APIs for coding assistance now have access to high-performing, on-premises options.

Looking ahead: The release of Dracarys signals ongoing advancements in AI-assisted coding and software development.

• As more models adopt similar fine-tuning techniques, we may see a general improvement in the coding capabilities of open-source LLMs.
• The competition between open-source and closed-source models for coding tasks is likely to intensify, potentially driving further innovation in the field.
• The integration of these advanced coding models into development workflows could reshape how software is created and maintained in the enterprise environment.