Student Researcher @google-deepmind. Prev. Visiting @Stanford | Tiger lab @Waterloo [email protected]
Repositories
Zhuofeng-Li/Zhuofeng-Li
Zhuofeng-Li/tmp-1
Zhuofeng-Li/llm-reranker
Zhuofeng-Li/EdgeBench
EdgeBench: Unveiling scaling laws of learning from real-world environments
Zhuofeng-Li/RL
Scalable toolkit for efficient model reinforcement
Zhuofeng-Li/Megatron-Bridge
Training library for Megatron-based models with bidirectional Hugging Face conversion capability
Zhuofeng-Li/llm-arch-table
Living comparison table of LLM architectural choices (norm, attention, MoE, positional encoding, and more) from the Original Transformer (2017) to frontier models (2026). Based on Harm de Vries's figure, Sebastian Raschka's Big LLM Architecture Comparison, and Tatsunori Hashimoto's Stanford CS 336 lecture.
Zhuofeng-Li/modded-nanogpt
NanoGPT (124M) in 90 seconds
Zhuofeng-Li/Megatron-LM
Ongoing research training transformer models at scale
Zhuofeng-Li/claude-code-settings
Zhuofeng-Li/harbor
Harbor is a framework for running agent evaluations and creating and using RL environments.
Zhuofeng-Li/CPRet
[NeurIPS'25] CPRet: A Dataset, Benchmark, and Model for Retrieval in Competitive Programming
Zhuofeng-Li/OpenPARL
Reproduction of Kimi K2.5 PARL Agent Swarm on WideSearch
Zhuofeng-Li/tmax
Training terminal-agents
Zhuofeng-Li/Server_set_up
Zhuofeng-Li/Agent.md
Zhuofeng-Li/QUEST
"QUEST: Training Frontier Deep Research Agents with Fully Synthetic Tasks"
Zhuofeng-Li/swarm-tasks
Zhuofeng-Li/ProRL-Agent-Server
Agentic RL on Any Harness at Scale
Zhuofeng-Li/slime
slime is an LLM post-training framework for RL Scaling.
Zhuofeng-Li/FrontierSmith
FrontierSmith, a new system that uses AI to synthesize open-ended coding problems at scale
Zhuofeng-Li/TEG-Benchmark
[NeurIPS 2024 🔥] TEG-DB: A Comprehensive Dataset and Benchmark of Textual-Edge Graphs
Zhuofeng-Li/frontier-cs
Zhuofeng-Li/swarm_example
Zhuofeng-Li/mac-transfer
My dotfiles
Zhuofeng-Li/DCI-Agent-Lite
Zhuofeng-Li/graph-mining-community-detection
CSCE 676 Final Project: Community detection in social networks using Louvain, structural features, and Node2Vec graph embeddings
Zhuofeng-Li/claude-code
An independent Python feature port of Claude Code, entirely rewritten from scratch. Educational Purpose only.
Zhuofeng-Li/OpenResearcher
OpenResearcher: A Fully Open Pipeline for Long-Horizon Deep Research Trajectory Synthesis