Projects Recently Starred by Experts

Starred 14 hours ago by

Jesse Clark(Cofounder of Marqo). Also starred by

Luis Capelo(Cofounder of Lightning AI),

Eric Zhang(Founding Engineer at Modal),

Pawel Garbacki(Cofounder of Fireworks AI), and

2 more.

sonic-moe by Dao-AILab

720

Accelerating Mixture-of-Experts (MoE) models

IO-aware and tile-aware optimizations for high-performance MoE.
Leverages CuTeDSL and Triton for state-of-the-art throughput.
Optimized for NVIDIA Hopper and Blackwell GPUs.
Built on Grouped GEMM kernels from QuACK/CUTLASS library.

Created 6 months ago

Updated 3 days ago

Starred 15 hours ago by

Zhuohan Li(Coauthor of vLLM). Also starred by

Vincent Weisser(Cofounder of Prime Intellect).

simply by google-deepmind

0.2%

535

Minimal JAX codebase for accelerating frontier AI research

Enables rapid iteration and implementation of novel LLM research ideas, optimizers, and training techniques.
Scalable JAX environment supporting LLMs and autoregressive models for human and AI agent-driven research.
Integrates Orbax for checkpointing and Grain for data pipelines, facilitating end-to-end research workflows.
Designed for AI agents to autonomously propose, run, and iterate on research experiments.

Created 1 year ago

Updated 1 week ago

Starred 15 hours ago by

Jesse Clark(Cofounder of Marqo). Also starred by

Jeff Hammerbacher(Cofounder of Cloudera),

Ross Taylor(Cofounder of General Reasoning; Cocreator of Papers with Code),

Yaowei Zheng(Author of LLaMA-Factory), and

3 more.

curator by bespokelabsai

0.3%

Synthetic data curation tool for post-training and structured data extraction

Generates synthetic data pipelines for model training or structured data extraction.
Supports structured outputs, async operations, caching, and fault recovery.
Integrates LiteLLM, vLLM, and batch APIs for diverse inference options.
Includes a viewer for monitoring data generation.

Created 1 year ago

Updated 1 week ago

Starred 15 hours ago by

Didier Lopes(Founder of OpenBB). Also starred by

Travis Fischer(Founder of Agentic) and

Justin Torre(Cofounder of Helicone).

open-knowledge by inkeep

149100.0%

AI-native markdown editor and LLM wiki

WYSIWYG editing for markdown, feeling like Notion or Google Docs.
Integrates LLMs (Claude, Codex) for AI-assisted content creation and editing.
Enables agentic search, knowledge graphs, and agent second brains via MCP.
Local, private, free knowledge base with Git sync and team sharing.

Created 3 weeks ago

Updated 15 hours ago

Starred 18 hours ago by

Elie Bursztein(Cybersecurity Lead at Google DeepMind). Also starred by

Pawel Garbacki(Cofounder of Fireworks AI) and

Yaowei Zheng(Author of LLaMA-Factory).

EverOS by EverMind-AI

12.2%

10k

Enterprise-grade AI memory system for persistent understanding

Enables AI to build understanding from past conversations, creating coherent narratives and living user profiles.
Features evidence-based perception with hybrid retrieval (RRF, semantic, keyword) and LLM-guided agentic recall.
Leverages multi-level memory structures (MemCells, episodes) and integrates with MongoDB, Elasticsearch, and Milvus.
Achieved 92.3% reasoning accuracy on the LoCoMo benchmark, demonstrating advanced AI understanding.

Created 8 months ago

Updated 1 day ago

Starred 1 day ago by

Jesse Clark(Cofounder of Marqo). Also starred by

Chip Huyen(Author of "AI Engineering", "Designing Machine Learning Systems"),

Eric Zhang(Founding Engineer at Modal),

Nathan Lambert(Research Scientist at AI2), and

8 more.

DeepGEMM by deepseek-ai

0.7%

CUDA library for efficient FP8 GEMM kernels with fine-grained scaling

Supports normal and Mix-of-Experts (MoE) grouped GEMMs on NVIDIA Hopper.
Employs CUDA-core two-level accumulation for FP8 tensor core imprecision.
Uses Just-In-Time (JIT) compilation for runtime kernel building.
Matches or exceeds expert-tuned libraries in performance across matrix shapes.

Created 1 year ago

Updated 1 day ago

Starred 1 day ago by

Jesse Clark(Cofounder of Marqo). Also starred by

George Hotz(Author of tinygrad; Founder of the tiny corp, comma.ai),

Wei-Lin Chiang(Cofounder of LMArena),

Luis Capelo(Cofounder of Lightning AI), and

3 more.

KernelBench by ScalingIntelligence

1.3%

Benchmark for LLMs generating GPU kernels from PyTorch ops

Evaluates LLM-generated CUDA kernels for correctness and performance (speedup vs. PyTorch).
Includes single-kernel ops, fused patterns, and full model architectures (MobileNet, VGG, Mamba).
Uses "fast_p" metric: fraction of correct kernels with >p speedup.
Provides scripts for evaluation, analysis, and baseline timing.

Created 1 year ago

Updated 3 months ago

Starred 1 day ago by

Jesse Clark(Cofounder of Marqo). Also starred by

Jeff Hammerbacher(Cofounder of Cloudera),

Pawel Garbacki(Cofounder of Fireworks AI),

Wing Lian(Founder of Axolotl AI), and

2 more.

autokernel by RightNow-AI

0.8%

Autonomous GPU kernel optimization for PyTorch

Optimizes PyTorch models by iteratively refining Triton or CUDA C++ GPU kernels.
Supports 9 core DL kernel types: matmul, attention, MLPs, and more.
Uses agent-based autoresearch, Amdahl's Law, and rigorous correctness verification.
Exports optimized kernels to HuggingFace Hub for seamless integration.

Created 3 months ago

Updated 3 months ago

Starred 1 day ago by

Jeremy Howard(Cofounder of fast.ai).

Awesome-Model-Merging-Methods-Theories-Applications by EnnengYang

0.5%

760

Survey of model merging methods, theories, applications for LLMs & MLLMs

Comprehensive overview of model merging techniques and taxonomy.
Explores applications in LLMs, MLLMs, and machine learning subfields.
Discusses continual learning, multi-task learning, and few-shot learning.
Highlights challenges and future research directions in model merging.

Created 2 years ago

Updated 3 days ago

Starred 1 day ago by

Max Howell(Author of Homebrew).

pixijs-skills by pixijs

8.6%

266

AI skills for 2D rendering agents

Master PixiJS v8 for WebGL, WebGPU, and Canvas graphics with AI coding agents.
Covers sprites, text, filters, custom rendering, and scene graph management.
Integrates with 40+ AI agents via skills CLI for seamless development assistance.
Optimizes performance, aids migration, and supports advanced rendering techniques.

Created 3 months ago

Updated 3 weeks ago

...