Paper collections of multi-modal LLM for Math/STEM/Code.

OCR and Voice Recognition Module: Effortlessly convert documents and audio into actionable text using advanced OCR engines and voice recognition technologies, featuring LLM correction and GPU acceleration—perfect for processing all kinds of hard data like math formula!

Complete mathematics curriculum for AI/ML/LLM - from foundations to research frontiers

One tiny lib for LLM token + cost math

This repository hosts the paper “LLM Based Math Tutoring: Challenges and Dataset”, along with the accompanying dataset. It explores the performance and challenges of Large Language Models (LLMs) in math tutoring scenarios, providing a benchmark dataset for evaluating LLM accuracy in educational contexts.

AISecOps (AI Security Operations) framework for deterministic verification of AI systems. QWED verifies LLM outputs using math, logic, and symbolic execution — creating an auditable trust boundary for agentic AI systems. Not generation. Verification.

An MCP Server for WolframAlpha's LLM API, able to return structured knowledge & solve math

Java 23, SpringBoot 3.4.1 Examples using Deep Learning 4 Java & LangChain4J for Generative AI using ChatGPT LLM, RAG and other open source LLMs. Sentiment Analysis, Application Context based ChatBots. Custom Data Handling. LLMs - GPT 3.5 / 4o, Gemini Pro 1.5, Claude 3, Llama 3.1, Phi-3, Gemma 2, Falcon 3, Qwen 2.5, Mistral Nemo, Wizard Math

Ongoing research project for code&math LLMs

A collection of recent open-source math datasets for training and evaluating Math LLMs

Created and enhanced a local LLM training system on Apple Silicon with MLX and Metal API, overcoming the absence of CUDA support. Fine-tuned the Llama3 model on 16 GPUs for streamlined solution of verbose math word problems. Result: a powerful, privacy-preserving chatbot that runs smoothly on-device.

A learning path for building AI systems. The project is a football match prediction system: data pipeline, Elo ratings, logistic regression, an LLM layer for match context, and a backtest against real bookmaker odds. Prerequisites are Python basics and high school maths.

Data and Code for ACL 2024 paper "DocMath-Eval: Evaluating Math Reasoning Capabilities of LLMs in Understanding Long and Specialized Documents"

🔬 The most atomic GPT-2 implementation in 265 lines of pure Python & CUDA. A bilingual "Rosetta Stone" for understanding LLM internals from scratch. No dependencies, just math and kernels.

Codebase for Math Neurosurgery: Isolating LLMs' Math Reasoning Abilities Using Only Forward Passes

Here we provide and collect many functions to generate math problem and step by step solutions for LLM training

An Advanced Basic Math Reasoning and Overthinking Evaluation Framework for LLMs

Official evaluation code for the U-MATH and μ-MATH benchmarks. These datasets are designed to test the mathematical reasoning and meta-evaluation capabilities of LLMs on university-level problems.

Code for "All for One: LLMs Solve Mental Math at the Last Token With Information Transferred From Other Tokens"

Implementing Supervised Fine-Tuning (SFT) and Reinforcement Learning (RL) for Qwen3 and DeepSeek-Math models. Includes experimental code, training logs, and insights on improving mathematical reasoning in LLMs.

re!think it. System prompt teaching LLMs to execute two core tasks: complex answers without hallucinations, and creative ideas without clichés. Written in math-like logic, which LLMs parse better than plain language. Built for mid-to-high complexity tasks, featuring a Bypass branch to execute simple prompts directly without added cognitive overhead

Description: PetroGPT focus on Gas/Oil Refinement/Chemistry specialized LLMs including WestSeverus-7B-DPO-v2 is a powerful multi-domain language model, combining Mistral-7B, FerdanoGPT (math-focused), and WestLake for scientific, mathematical, and programming tasks. Ranked #1 on Open LLM leaderboard, available on Hugging Face.

Ascend Flow (智流) is a multi-agent, adaptive learning system where specialized LLM tutors collaboratively plan, teach, generate scaffolded practice, and regulate motivation to keep learners in flow—bridging the gap from “understanding” to “solving” for coding interviews and foundational math.

No description available

VS Code Extension for WriteTex, write math equations and figures on tablet, sync to VS Code powered by context aware LLM OCR.

Online Markdown editor with advanced support for math equations, LLM-generated Math's Syntax, Syntax Highlighting, and diagrams.

LLM pipeline solution for Elementary Maths Solving (ZALO AI CHALLENGE 2023)

Agent stability, introspection & anti-drift framework for OpenClaw. Shannon entropy monitoring, confabulation detection, loop guards, principle-aligned growth tracking, and structured heartbeat decisions. Model-agnostic — works with Claude, GPT, Gemini, DeepSeek, or any LLM. Define your principles, the math enforces alignment.

Evaluation Code and Dataset for [ICML 2025] MATH-Perturb: Benchmarking LLMs' Math Reasoning Abilities against Hard Perturbations

“Semantic Latent Engineering (SLE) is a high-level framework for reasoning about meaning, intent and agents on top of LLMs. It combines math, cognitive architecture and practical patterns (SD, HDSA, ABC, CPP, MMOR) to move from prompt tricks to auditable, testable semantic engineering.