Search Results for "benchmark"

...Instead of manually tuning prompts or workflows, developers define high-level goals in a configuration file, and the system continuously modifies its own tools, orchestration, and logic based on benchmark performance. It operates through a loop of testing, analyzing failures, and refining the agent’s configuration to maximize a scoring metric. The framework uses a single-file agent harness combined with structured tasks and evaluation suites to guide optimization. It runs inside Docker for safe execution and reproducibility. This approach shifts agent development from manual design to automated optimization. ...

...The system operates in a ReAct-style loop where the agent profiles baseline implementations, writes CUDA code, compiles it in a sandbox, and iteratively refines performance. CUDA-Agent has demonstrated strong benchmark results, achieving high pass rates and significant speedups compared with compiler baselines such as torch.compile.

MiroThinker is an open-source deep research AI agent designed to perform complex reasoning, information gathering, and predictive analysis tasks. The system focuses on enabling long-horizon research workflows by allowing the agent to interact repeatedly with external tools, search systems, and data sources while refining its reasoning through iterative steps. Rather than simply generating responses from a single prompt, the agent performs structured multi-step reasoning processes that...

MiroFlow is a high-performance open-source framework designed for building intelligent AI agents capable of solving complex reasoning and research tasks. The system introduces a hierarchical architecture that organizes components into control, agent, and foundation layers, allowing developers to manage agent orchestration and tool interactions in a structured manner. One of the core innovations of MiroFlow is its use of agent graphs, which enable flexible orchestration of multiple sub-agents...

rLLM is an open-source framework for building and training post-training language agents via reinforcement learning — that is, using reinforcement signals to fine-tune or adapt language models (LLMs) into customizable agents for real-world tasks. With rLLM, developers can define custom “agents” and “environments,” and then train those agents via reinforcement learning workflows, possibly surpassing what vanilla fine-tuning or supervised learning might provide. The project is designed to...

Optimized Workforce Learning for General Multi-Agent Assistance in Real-World Task Automation. OWL (Optimized Workforce Learning for General Multi-Agent Assistance in Real-World Task Automation) is an advanced framework designed to enhance multi-agent collaboration, improving task automation across various domains. By utilizing dynamic agent interactions, OWL aims to streamline and optimize complex workflows, making AI collaboration more natural, efficient, and adaptable. It is built on...