Projects

Overview: This repository contains a curated list of research papers and resources on repository-level code generation and issue resolution, focusing on enhancing the efficiency and accuracy of software development processes. Key features: - Evaluating Code Slop in Long-Horizon Coding Agents - SkillForge: Self-Distilling Agents for Project-Specific Issue Resolution - LLM Agents Can See Code Repositories - Function-Aware Fill-in-the-Middle as Mid-Training for Coding Agent Foundation Models - Know Before Fix: QA-Driven Repository Knowledge Acquisition for Software Issue Resolution - Steer, Don't Solve: Training Small Critic Models for Large Code Agents - DeNovoSWE: Scaling Long-Horizon Environments for Generating Entire Repositories from Scratch - SWE-AGILE: A Software Agent Framework for Efficiently Managing Dynamic Reasoning Context - GALA: Multimodal Graph Alignment for Bug Localization in Automated Program Repair - Triage: Routing Software Engineering Tasks to Cost-Effective LLM Tiers via Code Quality Signals - ABTest: Behavior-Driven Testing for AI Coding Agents - Inside the Scaffold: A Source-Code Taxonomy of Coding Agent Architectures - Beyond Fixed Tests: Repository-Level Issue Resolution as Coevolution of Code and Behavioral Constraints - DebugHarness: Emulating Human Dynamic Debugging for Autonomous Program Repair - Yet Even Less Is Even Better For Agentic, Reasoning, and Coding LLMs - From SWE-ZERO to SWE-HERO: Two-Stage SFT for SWE-Bench Tech stack:

### Overview: Chatbot NER is an open-source framework designed specifically for Named Entity Recognition (NER) in text messages. Created by the Haptik team, this framework supports multiple Indian languages and their dialects, including English, Hindi, Gujarati, Marathi, Bengali, and Tamil. It uses common patterns and NLP techniques to extract entities from languages with sparse data, making it suitable for Conversational AI applications. ### Key Features: - **Supported Entities:** - **Time:** Detects time from text. - **Date:** Detects dates from text. - **Number:** Detects numbers and units from text. - **Phone Number:** Detects phone numbers from text. - **Email:** Detects email addresses from text. - **Text:** Detects custom entities in text strings. - **PNR:** Detects PNR codes from text. - **Regex:** Detects entities using custom regular expressions. ### Tech Stack: The framework is built using Python and Docker. ### Why It Matters: This framework is crucial for Conversational AI applications, providing a robust solution for entities extraction in various Indian languages and dialects. It helps in improving the accuracy and relevance of responses in real-time interactions, enhancing the overall user experience.

Overview: SynDiff is an official PyTorch implementation of the SynDiff model, described in the paper "Unsupervised Medical Image Translation With Adversarial Diffusion Models" by Muzaffer Özbey et al. The model is designed for unsupervised medical image translation tasks. Key features: - Official PyTorch implementation - Supports T1->T2 and T2->T1 translations in the IXI dataset - Supports T1->PD and PD->T1 translations in the BRATS dataset - Uses adversarial diffusion models for unsupervised image translation Tech stack: PyTorch, CUDA 11.2, ninja, Python 3.x Why it matters: SynDiff provides a powerful framework for unsupervised medical image translation, enabling researchers and practitioners to quickly test and evaluate new translation models without the need for paired training data.

### Overview: ReaL (ReaL: Efficient RLHF Training for LLMs with Parameter Reallocation) is a distributed system designed for efficient RLHF training with large language models (LLMs). It introduces a novel approach called parameter reallocation to dynamically redistribute LLM parameters across the cluster and adapt parallelization strategies during training. ReaL achieves significantly higher PPO training throughput compared to state-of-the-art open-source systems. ### Key Features: - **Parameter Reallocation**: Dynamically redistributes LLM parameters across the cluster to optimize parallelization and adapt to workload. - **Large-Scale Training**: Supports high-throughput generation with CUDAGraph and large-scale training with 3D parallelism. - **Memory-Efficient Training**: Enables memory-efficient training with parameter and optimizer offloading. - **Ease of Use**: Seamlessly integrates with HuggingFace checkpoints and inference frameworks like vLLM. No checkpoint conversion required. - **Versatile Configuration**: Offers versatile configuration customization with Hydra structured config. - **Customizable Algorithms**: Supports many commonly used RLHF algorithms, including DPO, PPO, RAFT, and more. Allows the addition of custom algorithms like [ReMax](https://github.com/openpsi-project/ReaLHF/tree/main/examples/new_algorithms/reinforce), [GRPO](https://github.com/openpsi-project/ReaLHF/tree/main/examples/new_algorithms/grpo), [Reference Model EMA](https://github.com/openpsi-project/ReaLHF/blob/main/examples/customized_exp/ppo_ref_ema

Overview: This repository contains the official code and documentation for the paper "Diffusion-based Time Series Imputation and Forecasting with Structured State Space Models," which was accepted by TMLR. The repository features a novel model architecture called $`SSSD^{S4}`$ that excels in handling diverse missing scenarios and datasets. Key features: - $`SSSD^{S4}`$ model architecture - Diverse algorithms combining diffusion and state-space models - State-of-the-art results across various time series scenarios Tech stack: Python Why it matters: This model offers significant improvements in time series imputation and forecasting, particularly in capturing long-term dependencies and handling missing data effectively across different datasets and scenarios. Repository: AI4HealthUOL/SSSD Primary language: Python GitHub stars: 335

### Overview: Abel is a state-of-the-art mathematical language model designed for generating mathematical expressions and reasoning. It has achieved remarkable performance across various benchmarks, including GSM8K, MATH, MathQA, SVAMP, and SCQ5K-EN, with an accuracy of over 80% and 29.46% respectively. ### Key Features: - **Model Performance**: Abel-7B-002 excels in mathematical datasets, achieving 80.44% accuracy on GSM8K and 29.46% on MATH. - **Generalization**: The model demonstrates strong generalization capabilities, performing well on out-of-domain datasets as well. - **Evaluation Metrics**: Abel-7B-002 outperforms other models across various benchmarks, including Qwen-7b, Mistral-7b, Yi-6b, and LLaMA2-7b, with scores ranging from 55.95% to 86.85%. ### Tech Stack: Abel is built using the Hugging Face framework, leveraging the LLaMA2 architecture for high performance and accuracy. ### Why It Matters: Abel is a significant advancement in the field of mathematical language models, offering unparalleled performance and generalization capabilities. Its ability to handle complex mathematical reasoning tasks makes it a valuable tool for researchers, educators, and anyone involved in mathematical research and education. ### GitHub Stars: The repository has been starred by 335 developers, indicating its growing interest and importance in the community. ###

Overview: This repository is a curated list of papers on misinformation research using large language models (LLMs), focusing on detection and verification methods. Key features: - **Cheap-fake Detection with LLM using Prompt Engineering** - **Faking Fake News for Real Fake News Detection** - **Bad Actor, Good Advisor: Exploring the Role of Large Language Models in Fake News Detection** - **Analysis of Disinformation and Fake News Detection Using Fine-Tuned Large Language Model** - **Detecting Misinformation with LLM-Predicted Credibility Signals and Weak Supervision** - **FakeGPT: Fake News Generation, Explanation and Detection of Large Language Model** - **Fighting Fire with Fire: The Dual Role of LLMs in Crafting and Detecting Elusive Disinformation** - **Language Models Hallucinate, but May Excel at Fact Verification** - **Clean-label Poisoning Attack against Fake News Detection Models** - **Rumor Detection on Social Media with Crowd Intelligence and ChatGPT-Assisted Networks** - **LLMs are Superior Feedback Providers: Bootstrapping Reasoning for Lie Detection with Self-Generated Feedback** - **Can Large Language Models Detect Rumors on Social Media?** - **TELLER: A Trustworthy Framework for Explainable, Generalizable and Controllable Fake News Detection** - **DELL: Generating Reactions and Explanations for LLM-Based Misinformation Detection** Tech stack: Python, Markdown, LaTeX, and various LLMs. Why it matters: This repository provides a comprehensive overview of the latest research in misinformation detection and verification

### Overview: This repository is a curated collection of papers and resources focusing on the latest advances in neuro-symbolic learning, specifically in the realm of large language models (LLMs). It aims to improve the reasoning and planning capabilities of LLMs through neuro-symbolic methods. ### Key Features: - **Neuro-Symbolic Frameworks**: Explore various neuro-symbolic frameworks and their applications. - **Symbolic to LLM**: Techniques for integrating symbolic reasoning with LLMs. - **LLM to Symbolic**: Methods for transferring knowledge from LLMs to symbolic representations. - **LLM plus Symbolic**: Techniques for enhancing LLMs with symbolic capabilities. - **Awesome Datasets & Benchmarks**: Resources for evaluating and benchmarking neuro-symbolic models. ### Tech Stack: Neuro-symbolic learning and reasoning. ### Why It Matters: This repository serves as a valuable resource for researchers, developers, and practitioners interested in advancing the capabilities of large language models through neuro-symbolic methods. It provides a comprehensive overview of the latest research, tutorials, and resources, making it an essential tool for anyone looking to stay updated on the cutting-edge developments in this field.

### Overview: This repository is a curated collection of state-of-the-art research papers, code, and resources related to hallucinations in Large Vision Language Models (LVLMs), also known as Multimodal Large Language Models (MLLMs). The main intention is to provide a platform where researchers can access and contribute to the field of hallucination in LVLMs. ### Key Features: - **Evaluation Benchmark**: Provides a comprehensive benchmark for evaluating hallucinations in LVLMs, covering various aspects such as object hallucination, attribute hallucination, and description hallucination. - **Detection**: Offers tools and techniques for detecting hallucinations in LVLMs, including fine-grained annotations and direct preference optimization. - **Mitigation**: Provides strategies and techniques for mitigating hallucinations in LVLMs. - **Survey**: A survey of recent research papers and studies on hallucinations in LVLMs. ### Tech Stack: The repository uses Markdown, GitHub Pages, and other open-source technologies to organize and present the information. ### Why It Matters: The repository aims to provide a structured and constructive way for researchers to access and contribute to the field of hallucinations in LVLMs. It serves as a valuable resource for anyone interested in advancing the understanding and application of hallucinations in large language models. ### Repository: NishilBalar/Awesome-LVLM-Hallucination - **GitHub Stars**: 335 - **GitHub Description**: "up-to-date curated list of state-of-the-art Large vision language models hallucinations research work, papers & resources" - **Primary

### Overview: Whispering Tiger UI is a native Go/Fyne UI for Windows and Linux, providing real-time transcription, translation, and text-to-speech capabilities. It supports local AI processing, enabling offline use after model downloads. ### Key features: - **Native UI for Windows and Linux** - **Local AI processing**, usable offline after downloading the selected models - **CPU and NVIDIA CUDA** for compatible models; **AMD, Intel and NVIDIA Vulkan** for audio.cpp STT/TTS ([runtime limits](doc/hardware-support.md)) - **Access to all Whispering Tiger features**, including transcription, translation, text-to-speech, in-game image recognition, and browser and VRChat integration via WebSockets or OSC - **Desktop audio capture** through WASAPI loopback on Windows or PulseAudio/PipeWire monitor sources on Linux - **Audio routing** for additional sources, translation and TTS output ([audio setup](doc/audio-config.md)) - **Save** and **load** configurations - **Preview** if your selected Audio devices are working - **Plugin** support for additional features, including a Realtime Subtitles Plugin, Text2Speech Plugins, Emotion Prediction Plugin, and more ### Tech stack: Go/Fyne ### Why it matters: Whispering Tiger UI offers a powerful and versatile solution for live transcription, translation, and text-to-speech, making it an essential tool for users looking to enhance their productivity and communication skills. ### Repository: Sharrnah/whispering-ui ### GitHub stars: 334 ### GitHub

### Overview: This project is a code implementation of a paper titled "Evaluation of deep neural networks for traffic sign detection systems". It evaluates various deep learning models for traffic sign detection, comparing their performance metrics such as mean average precision (mAP), memory allocation, running time, and parameters. The paper focuses on adapting pre-trained models for traffic sign detection using transfer learning techniques. ### Key Features: - **Pretrained Models**: Available TFRecords of the GTSDB dataset and pre-trained models from Google Drive. - **Experiment Results**: Detailed evaluation metrics and comparisons of various models. - **Running on New Images**: Instructions for running the model on new traffic sign images. - **Training Your Model**: Detailed steps for training your own model using the provided pre-trained models. - **Evaluating Your Model**: Guidance on evaluating the performance of your trained model. - **Scripts**: Scripts for creating GTSDB TFRecords. ### Tech Stack: The project is implemented using TensorFlow and is based on the Tensorflow Object Detection API and darkflow repositories. ### Why It Matters: This project provides valuable insights into the performance and effectiveness of different traffic sign detection models, particularly in the context of traffic sign detection systems. The findings can be crucial for researchers and developers looking to improve traffic sign detection algorithms.

### Overview: NLPGNN is a comprehensive toolbox designed for natural language processing (NLP) tasks, focusing on graph neural network (GNN) models. It leverages advanced techniques such as BERT, ALBERT, and GPT2, along with various GNN methods like GCN, GAN, GIN, and GraphSAGE, to simplify the implementation of complex NLP models. ### Key Features: - **BERT, ALBERT, and GPT2 Integration**: Utilizes TensorFlow 2.0 layers for efficient NLP tasks. - **Graph Neural Network (GNN) Models**: Implements GCN, GAN, GIN, GraphSAGE, TextGCN, and TextSAGE for text classification and graph classification. - **Model Implementation**: Provides a simple framework to integrate various GNN models, making it easier to experiment with different architectures. ### Tech Stack: - Python ### Why It Matters: NLPGNN simplifies the implementation of complex NLP models by providing a unified framework that integrates advanced GNN techniques. This toolbox is particularly useful for researchers and developers looking to experiment with GNN models in NLP without the need for extensive custom code. ### Repository: kaiyinzhou/NLPGNN - **Primary Language**: Python - **GitHub Stars**: 335 - **GitHub Description**: This repository is dedicated to the field of natural language processing, focusing on GNN models. It includes models like BERT, ALBERT, GPT2, TextCNN, GCN, GAN, GIN, Graph

Overview: This repository contains code and pre-trained models for Deep Semantic Role Labeling (Deep SRL), a technique for automatically identifying and labeling the roles played by verbs in sentences. The models are based on the research presented in "Deep Semantic Role Labeling: What Works and What's Next" by He et al., which is available for download. Key features: - Pre-trained models for CoNLL-2005 and CoNLL-2012 datasets - Code for training and using the Deep SRL model - Scripts for fetching and processing required data, including CoNLL05 and CoNLL2012 datasets - An interactive console for quick experimentation with the model Tech stack: Python Why it matters: The Deep SRL model is significant because it leverages deep learning techniques to automatically identify and label the roles of verbs in sentences, providing a powerful tool for natural language processing tasks. The pre-trained models and scripts make it easy to replicate and evaluate the results on the CoNLL datasets, which are widely used benchmarks in the NLP community. Repository: luheng/deep_srl GitHub stars: 335

Overview: This package provides a simple and efficient way to interact with the OpenAI API using PHP, including all endpoints and models. Key features: - ChatGPT (GPT-3.5 & GPT-4) - GPT-3.5 - GPT-4 - Codex - DALL·E - Whisper - Fine-Tuning - Embeddings - Moderation Tech stack: PHP Why it matters: This package simplifies the integration of OpenAI into PHP applications, providing fully typed Data Transfer Objects (DTOs) and IDE autocomplete support, making it easier and more reliable for developers. Repository: tectalichq/public-openai-client-php Primary language: PHP GitHub stars: 335

### Overview: HyMPS is a platform-independent software suite designed for advanced audio/video contents production. ### Key Features: - **Audio Production Tools:** Includes AI-based, DAWs, DSPs, editors, effects, libraries, sequencers, synthesizers, recorders, tools, treatments, engines, and programming. - **Video Production Tools:** Covers AI-based, editors, libraries, tools, processing, pan tilt zoom, and RTSP. ### Tech Stack: HyMPS is built using open-source technologies and frameworks. ### Why It Matters: HyMPS provides a comprehensive solution for creating high-quality audio and video content, making it an essential tool for professionals in the field.

### Overview: Qwen3 Fine-tuning: Medical R1 Style Chat is a project that fine-tunes the Qwen3 model for medical reasoning tasks. The project includes a detailed README that outlines the model, data, and training process, as well as a comprehensive tutorial and training logs. The project is designed to be used in medical settings to provide accurate and relevant responses to medical inquiries. ### Key Features: - **Model**: Fine-tuned Qwen3 model - **Data**: Delicate medical dataset - **Training**: Full parameter and LoRA micro-tuning - **Inference**: Inference capabilities for both full parameter and LoRA models - **Documentation**: Comprehensive documentation including a tutorial and training logs ### Tech Stack: - Python ### Why it Matters: This project is crucial for medical professionals and researchers as it provides a fine-tuned model for medical reasoning, enhancing the accuracy and relevance of responses to medical inquiries. The use of a detailed README and comprehensive training logs ensures that the model is well-documented and can be effectively used in real-world applications. ### Repository: - Repository: Zeyi-Lin/Qwen3-Medical-SFT - GitHub Stars: 334 - GitHub Description: Qwen3 Fine-tuning: Medical R1 Style Chat ### Installation: ```bash pip install -r requirements.txt ``` ### Data Preparation: The project automatically downloads, preprocesses, and partitions the dataset to generate `train.jsonl` and `val.jsonl` files. ### Training: - **Full Parameter Micro-T

Overview: This repository provides quick-start examples for OptaPlanner, a powerful AI optimization library. Ideal for developers looking to integrate AI optimization into their Java projects, these quick starts cover various use cases across multiple technologies. Key features: - Real-world use cases - Multiple technology integrations - Step-by-step guides Tech stack: Java Why it matters: These quick starts offer a practical way to incorporate AI optimization into Java projects, showcasing the library's capabilities across different technologies and use cases.

### Overview: DeepWorks is a collection of Deep Learning projects and resources created by Avkash Chauhan. It serves as a comprehensive guide for developers interested in the field of artificial intelligence and machine learning. ### Key Features: - **Deep Learning Projects**: A variety of projects and resources to enhance understanding and practical application of deep learning techniques. - **Python Data Processing**: A course for budding Python developers focusing on data processing and machine learning. - **Gradio Tabbed UI**: A guide on developing a tabbed user interface using Gradio. - **Python Module Packaging**: Steps to create, package, and install a Python module. - **Rust by ChatGPT**: A tutorial on Rust programming language, inspired by ChatGPT. ### Tech Stack: - Jupyter Notebook ### Why It Matters: DeepWorks provides a valuable resource for developers looking to enhance their skills in deep learning, Python programming, and other related technologies. The projects and tutorials are designed to be accessible and practical, making it an excellent guide for beginners and experienced developers alike. ### Repository: - Repository: avkash/DeepWorks - GitHub Stars: 334 - GitHub Description: A collection of Deep Learning projects and resources - GitHub URL: [https://github.com/prodramp/DeepWorks](https://github.com/prodramp/DeepWorks) ### Developer Program: - **Data Processing course in Python for budding Python Developers - 4 Parts** - **Developing Gradio Based Tabbed UI - 2 Parts** - **Creating Python Module, Packaging it as Wheel

### Overview: This repository contains the implementation of SE3-Transformers, a model designed for equivariant self-attention in PyTorch. It is specifically geared towards integration with the Alphafold2 project and other applications in drug discovery. ### Key Features: - **SE3-Transformers Implementation**: The repository provides a complete implementation of SE3-Transformers, including all necessary components for self-attention and equivariance. - **Pytorch Integration**: The code is written in PyTorch, making it compatible with various deep learning frameworks. - **Edge Information**: The model can handle edge information by passing additional keyword arguments during initialization, allowing for the inclusion of bond types or other edge features. ### Tech Stack: - Python - PyTorch ### Why it Matters: - **Integration with Alphafold2**: This model is essential for replicating the results of Alphafold2, which is crucial for drug discovery and molecular modeling. - **Equivariance**: The model ensures that the attention mechanism respects the symmetries of the system, making it particularly useful in applications where symmetry is important. - **Edge Information Handling**: The ability to incorporate edge information is a significant advantage, allowing for more realistic and context-aware modeling in molecular structures. ### GitHub Stars: - 334 ### Example Usage: ```python import torch from se3_transformer_pytorch import SE3Transformer # Example usage in Alphafold2 model = SE3Transformer( dim = 64, depth =

Overview: Agentic World Modeling is a comprehensive framework that explores the foundations, capabilities, laws, and beyond of human agency in the digital age. This repository serves as a curated collection of 1,000+ papers and benchmarks, organized by sections in reverse chronological order, providing a structured overview of the field's progress. Key features: - Comprehensive collection of papers and benchmarks - Organized by sections in reverse chronological order - Available in a public Notion database - Released under the MIT License - Cited in the research community Tech stack: TypeScript Why it matters: Agentic World Modeling offers a foundational understanding of human agency in digital environments, contributing to advancements in artificial intelligence, robotics, and related fields. Repository: matrix-agent/awesome-agentic-world-modeling GitHub stars: 330 GitHub description: Agentic World Modeling: Foundations, Capabilities, Laws, and Beyond