Juzheng Zhang

flash-bon. (ECCV 2026): Official code for Flash-BoN: Instant Drafts for Inference-Time Scaling in Diffusion Models

streaming. Code for the paper Multi-Stream LLMs: Unblocking Language Models with Parallel Streams of Thoughts, Inputs and Outputs

logit-fusion. Learning from Mixed Rollouts: Logit Fusion as a Bridge Between Imitation and Exploration

scalable-curvature. Code for Dayal Kalra's research internship on scalable curvature measures for neural networks.

retrofitting-recurrence. Teaching Pretrained Language Models to Think Deeper with Retrofitted Recurrence

nanochat. The best ChatGPT that $100 can buy.

nanoGPT. The simplest, fastest repository for training/finetuning medium-sized GPTs.

awesome-RLVR-boundary. A curated list of resources on Reinforcement Learning with Verifiable Rewards (RLVR) and the reasoning capability boundary of Large Language Models (LLMs).

rl-injector. Official release of code for the paper RL is a hammer and LLMs are nails A simple RL approach to stronger prompt injection attacks

refusal-tokens. This is the official repo for "Refusal Tokens: A Simple Way to Calibrate Refusals in Large Language Models"

zero. PyTorch Implementation of Zero-Shot Vision Encoder Grafting via LLM Surrogates [ICCV'25]

LoRI. [COLM 2025] LoRI: Reducing Cross-Task Interference in Multi-Task Low-Rank Adaptation

recurrent-pretraining. Pretraining and inference code for a large-scale depth-recurrent language model

HL-GNN. [KDD 2024] "Heuristic Learning with Graph Neural Networks: A Unified Framework for Link Prediction"