orchestra-research

Enhances PyTorch training scripts with FSDP2 for efficient model training across multiple GPUs, optimizing memory usage and performance.

Facilitates distributed training orchestration for machine learning models across clusters, enhancing scalability and efficiency.

Evaluates NVIDIA Cosmos Policy in LIBERO and RoboCasa environments, optimizing robot manipulation and profiling GPU performance.

Facilitates fine-tuning and serving of Physical Intelligence OpenPI models for robotic policy inference using JAX or PyTorch.

Guides researchers through structured ideation frameworks to discover impactful research directions and explore new problem spaces.

Compiles diverse research inputs into structured, machine-executable knowledge packages, enhancing research artifact creation.

Conducts comprehensive semantic epistemic reviews of research artifacts, scoring key dimensions to enhance research quality.

Guides the evolution and optimization of AI agents using LLM-driven algorithms for enhanced performance and adaptability.

Fine-tunes OpenVLA-OFT for robot action generation using advanced techniques like LoRA adaptation and continuous action heads.

Generates high-quality figures for ML papers by extracting data and relationships, utilizing Gemini for diagrams and matplotlib/seaborn for charts.

Assists in writing publication-ready ML/AI papers for top conferences, ensuring accurate citations and structured arguments.

Generates conference presentation slides from research papers, producing Beamer PDFs and editable PPTX with speaker notes.

Provides a comprehensive guide for writing systems papers with structural blueprints, templates, and venue-specific checklists.

Applies cognitive science frameworks to enhance creative thinking for novel research ideation in computer science and AI.

Records research provenance by extracting key events from coding sessions, ensuring an auditable trace of project evolution.

Guides users in tracking ML experiments with SwanLab, offering open-source run tracking and visualization for various ML frameworks.

Provides battle-tested PyTorch training recipes for various AI domains, optimizing model training and fine-tuning processes.