Third-year undergraduate in Network Engineering at Guangzhou University. I build tools that audit deployed AI systems, run preregistered empirical studies in RL, and contribute a small number of fixes upstream in depth — rather than scattering patches across many repos.
Applying for PhD programs (Fall 2027) in Embodied AI / Reinforcement Learning / Robotics.
Academic Homepage | Blog | CV | ldz@e.gzhu.edu.cn
TransitTruth — open-source auditing tool that verifies whether an AI API relay actually serves the advertised model, using statistical behavioral fingerprints (chi-square / KS tests, Bayesian updating). Includes a zero-install web demo and an extended technical report validating same-family discrimination (gpt-4o vs gpt-4o-mini, TVD up to 0.90, 91.3% in-distribution resubstitution posterior — self-consistency on the baseline's own reference, not held-out accuracy). Complements single-token fingerprinting work by focusing on fine-grained same-family distinctions. Tech Report (PDF) · Live Demo
wm-audit — preregistered single-machine CPU audit suite that disentangles world-model perceptual fidelity (next-frame MSE / SSIM / NLL, plus fixed-filter stand-ins for LPIPS / rollout FVD — deterministic random-conv filter banks, not learned VGG/AlexNet backbones) from control utility (whether a planner can actually solve the task). A reduced pilot on one env (pixel_nav, 3 seeds × 3 capacities, 12k steps) found: pixel fidelity weakly predicts closed-loop return (H1 supported, |ρ|=0.32<0.4); control-relevant diagnostics did not (H2 not supported, ρ=0.26<0.6); INT8 compression did not degrade calibration faster than raw pixel fidelity (H3 not supported, n_pairs=9). This is a CPU reduced pilot, not the locked 200k-step × 5-seed × 2-env main run.
embodied-offpolicy-study — a Soft Actor-Critic implementation written from scratch in PyTorch (no Stable-Baselines3), run as a preregistered three-arm controlled study on Hopper-v4 (CPU, 120k env steps, 3 seeds/arm): Baseline (reward scale 1.0) vs RewardScaled (0.1) vs FixedAlpha (α fixed at 0.2). Welch p: Baseline–FixedAlpha 0.59, Baseline–RewardScaled 0.95; ANOVA F=0.22, p=0.81 — no statistically distinguishable effect detected (n=3). At 60k the reward-scaled arm looked ~3× better; extending to the preregistered 120k budget and adding the fixed-temperature control arm overturned that apparent gain as early seed noise.
rl-deploy-bench — cross-platform RL deployment and benchmarking toolkit. Export SB3 policies to ONNX/TorchScript, build TensorRT engines with FP16/INT8 quantization, and benchmark latency/throughput/accuracy across GPU and CPU from one config-driven CLI with auto-generated HTML reports. Jetson/edge is a target deployment path, not yet measured on device.
Merged
| Project | PR | Contribution |
|---|---|---|
| microsoft/TextWorld | #377 | Fixed env.step crash on multi-command input (non-greedy score parsing + unit tests) |
| shmuma/ptan | #57 | Fixed Gymnasium API compatibility in experience sources |
| learnsyslab/gym-pybullet-drones | #319 | Replaced print() error statements with module logging; converted exit() calls to raised exceptions |
| Eclectic-Sheep/sheeprl | #340 | Added CONTRIBUTING guide and Code of Conduct |
Under review (not yet merged) — selected code PRs in upstream RL/ML infrastructure
| Project | PR | Contribution |
|---|---|---|
| thu-ml/tianshou | #1315 | Fixed pickled policy restore under PyTorch 2.6 weights_only default |
| ContinualAI/avalanche | #1701 | Fixed MASPlugin raising on zero loss (falsy check) |
| uoe-agents/epymarl | #92 | Fixed registration of multi-digit versioned PettingZoo envs |
| vllm-project/vllm | #57400 | Fixed system_fingerprint=null emitted under fingerprint-mode=none |
| huggingface/datasets | #8633 | Accepted pathlib.Path in load/save_to_disk |
| huggingface/diffusers | #14796 | Fixed undefined attribute in TangentialClassifierFreeGuidance |
| google/brax | #676 | Fixed EpisodeWrapper metrics accumulation with action_repeat |
| AI4Finance-Foundation/ElegantRL | #488 | Fixed argmax→max in get_cumulative_rewards |
| Farama-Foundation/PettingZoo | #1465 | Fixed agent selector reinit in tictactoe reset |
| MichalBortkiewicz/JaxGCRL | #65 | Fixed XML path resolution in arm manipulation envs |
| pandas-dev/pandas | #68968 | Fixed to_html href escaping with render_links + escape |
Smaller docs / community-health PRs (LICENSE, CONTRIBUTING, Code of Conduct, CI setup) across a number of smaller RL/ML projects, plus a typo fix in FluxVLA.
- Fixing a bug in Microsoft TextWorld — one line of regex and the multi-command parsing behind it: root-cause writeup of merged PR #377. A greedy regex with DOTALL stitched two state blocks into one when a single line carried multiple commands.
- CI passed green but the package wouldn't import: numpy and scipy missing from requirements: postmortem on a dependency-declaration bug found while running the TransitTruth audit pipeline — tests on a dirty machine passing says nothing about a clean install.
RL / Robotics — PyTorch · Stable-Baselines3 · Gymnasium · MuJoCo Deployment — ONNX · ONNX Runtime · TensorRT · TorchScript · INT8/FP16 Systems — Python · C++ · Linux · Docker · Git · CI/CD · CUDA