Symbolic step verification for logical reasoning in LLMs using constraint propagation.

Decode-aware prefill-stage KV selection for Multimodal LLMs

A reflective RL framework for generating high-performance CUDA kernels using structural and performance rewards.

Unified scientific multimodal foundation model for cross-domain scientific understanding and generation.

Lightweight virtualization technology for running OCI images as microVMs.

A comprehensive synthetic OCR dataset for the Persian language with 343k image-text pairs.

A hierarchical optimization framework for trillion-parameter MoE model post-training on Ascend NPU infra.

A profiler-guided, verifier-in-the-loop workflow for repository-level code optimization.

A minimal context management framework built around programmatic memory for LLM agents.

Code and data for statistically grounded sparse-feature interventions for LLM activation-space control.

Implementation of Exploratory Sampling (ESamp) for boosting Pass@k efficiency in reasoning models.

Framework for generalized physical world modeling with vision-language agents.

Language-conditioned Meta-Reinforcement Learning framework for efficient task adaptation.

Open-source framework for failure observability, attribution, and recovery in LLM agents.

Static Orthogonal-Subspace Low-Rank Adaptation for parameter-efficient fine-tuning of LLMs.

An evolutionary algorithm guiding LLMs to generate executable PINN configurations.

An infra layer for managing open-source apps and running coding agents.

An agentic CLI tool for coding and programming workflows.

A fast, memory-efficient pure-Transformer video decoder for real-time latent-to-pixel streaming.

Embodied Action Memory from Egocentric Videos for Reasoning and Planning