🚀 New model alert! Introducing "agentica-org_deepcoder-14b-preview"! A code reasoning LLM fine-tuned for long context lengths, achieving 60.6% Pass@1 accuracy on LiveCodeBench. Try it out with `local-ai run agentica-org_deepcoder-14b-preview` #LocalAI #NewModel #CodeReasoning

LLMs demonstrate impressive code generation abilities, but do they truly understand execution flow? In our paper, CoCoNUT, we evaluate seven SOTA LLMs on their ability to trace code execution, revealing significant limitations. #LLMs #CodeReasoning

🚀 Exciting news from Meta AI! A new benchmark, CRUXEval, is introduced to assess code language models' capability in code understanding and execution. Dive into the details here: ift.tt/MzbmXta #MetaAI #CRUXEval #CodeReasoning