yu_wang

Logic-RL-Lite. Lightweight replication study of DeepSeek-R1-Zero. Interesting findings include "No Aha Moment", "Longer CoT ≠ Accuracy", and "Language Mixing in Instruct Models".

Awesome-LLM-Interview-Questions-and-Answers. 大模型算法工程师、大模型Agent开发工程师面试：常见题目和答案

DeepEnlighten. Pure RL to post-train base models for social reasoning capabilities. Lightweight replication of DeepSeek-R1-Zero with Social IQa dataset.

byte_pair_encoding_BPE_subword_tokenization_implementation_python. Byte-Pair Encoding (BPE) (subword-based tokenization) algorithm implementaions from scratch with python

Reinforcement-Calibration-SimCSE. Reinforcement Calibration SimCSE, combining contrastive learning, artificial potential fields, perceptual loss, and RLHF to achieve improved Semantic Textual Similarity (STS) embeddings. PyTorch-based implementations of PerceptualBERT and ForceBasedInfoNCE, along with fine-tuning capabilities via RLHF and evaluation using SentEval.

Cheat-Sheet-for-NVIDIA-Certified-Associate-Generative-AI-LLMs-NCA-GENL-. A comprehensive cheat sheet and study guide for the NVIDIA Certified Associate - Generative AI LLMs (NCA-GENL) exam. Covers key concepts, the NVIDIA AI ecosystem, and includes both English and Chinese versions.

Sudoku4LLM. Sudoku4LLM is a Sudoku dataset generator for training and evaluating reasoning in Large Language Models (LLMs). It offers customizable puzzles, difficulty levels, and 11 serialization formats to support structured data reasoning and Chain of Thought (CoT) experiments.

awesome-tips-for-economic-phd. Collection of resources for economists and economics PhD students, including tips for academic writing, research, publishing, job market preparation, referee reports, and more.

event-timeline-generation-olympics. A toy system for generating event timelines from social media data, specifically focusing on the Olympic Game medalist events.

clustering_algorithm_implementation_python. Clustering algorithm implementaions from scratch with python (k-means, EM-GMM, mean-shift, agglomerative)

Stereo-Matching-Disparity-Maps. Comparative analysis of stereo matching techniques for disparity map reconstruction, including traditional block matching, OpenCV StereoSGBM, and PSMNet.

awesome-eye-tracking-economics-business. Collection of materials and papers on eye tracking in field of economics and business, covering the basics of eye tracking technology and its application to visual attention studies.

Awesome-Surveys-of-LLM4EDA. A curated list of surveys, reviews, and awesome lists on the application of Large Language Models (LLMs) and AI for Electronic Design Automation (EDA) and chip design.

Learn_Transformer_from_Easy_to_Hard. 一个为初学者设计的 Transformer 分阶段学习项目。通过由简到难的四个阶段，从零开始用 PyTorch 构建完整的 Transformer 模型，深入理解其核心模块与工作原理。

awesome-VR-AR-economics-business. Collection of papers on Virtual Reality (VR) and Augmented Reality (AR) applications in field of economics and business.

Awesome-Satellite-Data-in-Finance-and-Economics. Curated list of surveys and papers on the applications of satellite data (satellite imagery) in finance and economics.

awesome-cryptocurrency-blockchain-economics-business. Curated list of papers on cryptocurrency and blockchain from top Economics, Finance, and Information Systems journals.

DolbyUUU.

awesome-model-pricing-and-model-valuation. A curated list of academic papers focused on AI model pricing, model valuation (on model marketplaces).

regression_algorithm_implementation_python. Regression algorithm implementaion from scratch with python (OLS, LASSO, Ridge, robust regression)

Deep-Clustering-Economic-Preferences. Economic preference clustering analysis using generative and deep learning models, including Gaussian Mixture Models (GMM), Wishart Mixture Models (WMM), and Variational Deep Embedding (VaDE).

Virtual-Reality-based-Balloon-Analogue-Risk-Task-VR-BART. A Virtual Reality-based Balloon Analogue Risk Task (VR-BART) built with Unreal Engine 5 to study risk-taking behavior in immersive environments.

hangman-solver-game-dynamics. A game-dynamics-aware Hangman solver leveraging fine-tuned BERT, weighted N-grams, and information entropy. Achieves a 74.4% success rate on a disjoint dataset.