Tag
The paper proposes Generalized Agent Iteration (GAI) as a formal framework unifying iterative policy improvement and recursive self-improvement, defining agents as modifiable components and distinguishing cases based on external evaluation and internal improvement.
This paper introduces an open modular transformer-based framework for detecting and structuring social tipping point evidence in climate documents, utilizing models like DistilBERT, RoBERTa, Mistral 7B, and LLaMA 3.2 3B, with evaluation on expert-labeled benchmarks.
Mycelium is an open-source, event-driven platform enabling AI agents to collaborate in a real-time virtual office, featuring WebSocket streaming, SQLite persistence, and production-grade guardrails for developers.
PlannerForge introduces an LLM-agent framework for unified scenario-based testing in autonomous driving, showing improvements over existing methods with both commercial and open-source models.
MobileVLA-R1 2.0 is an RL-enhanced vision-language-action framework that couples structured reasoning with mobile robot control, achieving improvements in long-horizon instruction following on real-world platforms.
SGLang has added deployment recipes for DeepSeek-V4-Flash-Vision and DeepSeek-V4-Flash-0731 models on 2x DGX Spark hardware, with support for various configurations and optimizations.
PyTorch 2.14 release introduces fault tolerance as a first-class c10d concept with in-place process-group reconfiguration, alongside features like NVGEMM kernels, a new nccl2 backend, and native linear algebra for Apple Silicon.
LangChain announces updates to its MCP support, including the new stateless MCP spec, to enhance AI framework capabilities.
This paper introduces a control-data flow separation framework to stabilize prompt optimization in multi-agent LLM systems by decoupling execution protocols from language content, achieving 100% protocol validity while enhancing task performance.
InsightSR is a framework that leverages Large Language Models to refine the search space for symbolic regression, improving accuracy and physical consistency through iterative semantic and structural guidance.
Pipecat v1.8.0 is released with major updates including support for Gemini 3.5 Transcribe, improved error handling, faster pipeline startup, and various new integrations.
Headlong is an open-source microharness that enables persistent agency in AI agents, allowing them to continuously think and self-guide even without external input.
RomeAILab has open-sourced Rome, an agentic OS that enhances AI agents by scaling the environment with better tools, memory, and workflows for human-agent collaboration.
A patch release for DSPy includes CodeInterpreter fixes, MCP 2.0 support, GEPA bump, and improvements to errors and callbacks.
EnvHarness is a programmable wrapper framework that dynamically adapts static environments to enhance LLM agent training, outperforming existing methods by providing superior optimization signals for reinforcement learning across multiple benchmarks.
KerasFormers is a collection of pretrained models integrated into Keras 3, designed to streamline deep learning model deployment for developers.
Edward Yang will be a keynote speaker at PyTorchCon North America, discussing the evolution and future of PyTorch. The event is set for October 20-21 in San Jose, CA.
DeepSeek Harness is an AI framework where models, tools, and other components are plugins, with a curated ecosystem available on GitHub for developers.
LEGO-RL presents a framework to bridge native coding-agent harnesses with scalable policy-gradient reinforcement learning, improving performance on benchmarks like SWE-bench.
DeepSeek Harness is an open-source GitHub project with over 114k stars, offering a plugin-based architecture where models, tools, and configurations can be swapped at the config layer.