AI Trending

Project directory

Projects

Open-source AI projects tracked by the leaderboard, ordered by GitHub stars.

Go 5 stars

gpustack/gpustack-operator

Manage heterogeneous accelerators, vGPU slicing, shared KV cache, model delivery, and inference workloads.

### Overview: GPUStack Operator enables the management and utilization of heterogeneous accelerators on Kubernetes, including GPUs, NPUs, and other accelerators. It provides features such as device discovery, RDMA networking, topology placement, shared cache, model delivery, and model deployment, making it a comprehensive solution for accelerating machine learning workloads. ### Key Features: - **Heterogeneous Devices**: Discover and allocate whole devices, shared devices, logical slices, or hardware partitions. - **RDMA Networking**: Provide RDMA network interfaces alongside accelerators. - **Topology**: Place related workloads within the requested network or location domain. - **KV Cache**: Share inference cache across workloads. - **Model Delivery**: Fetch, verify, and cache model weights on nodes. - **Model Deployment**: Run model serving replicas with routing and optional prefill/decode roles. - **GPU Instances**: Enter accelerator-backed container workspaces over SSH. ### Tech Stack: Go ### Why it Matters: GPUStack Operator simplifies the management and utilization of accelerators on Kubernetes, making it easier to deploy and scale machine learning workloads.
Jupyter Notebook 5 stars

guy1998/1998-prompts

Fine tuned LLaMA model for prompt enhancing

Overview: This project fine tunes a LLaMA model for prompt engineering, addressing the lack of time and resources for personal prompt enhancement. Key features: - Fine tunes a LLaMA model for prompt engineering - Uses gradient accumulation for training stability - Supports quantization and inference for broader usage Tech stack: Jupyter Notebook Why it matters: The project aims to provide a quick and efficient solution for prompt engineering, addressing the need for prompt enhancement without the limitations of existing solutions. Repository: guy1998/1998-prompts Primary language: Jupyter Notebook GitHub stars: 5 GitHub description: Fine tuned LLaMA model for prompt enhancing
Python 5 stars

harsha-gouru/gpu-inference-playground

Benchmark LLM inference techniques on GPU — KV cache, quantization, FlashAttention, speculative decoding, continuous batching. Runs on H100/MI300X via Modal/RunPod.

### Overview: Benchmark and experiment with LLM inference techniques on GPU. Run a script, get results, learn. ### Key features: - KV cache optimization - fp16 vs int8 vs int4 tradeoff - Throughput scaling with batch size - FlashAttention2 vs SDPA vs eager attention - Draft+verify speculative decoding - Continuous batching for vLLM models ### Tech stack: Python ### Why it matters: Learn and benchmark different LLM inference techniques on GPU, including KV cache optimization, fp16 vs int8 vs int4 tradeoff, batch size scaling, FlashAttention2 vs SDPA vs eager attention, speculative decoding, and continuous batching for vLLM models.
Jupyter Notebook 5 stars

hassanzadehmahdi/Sarcasm-detection-in-social-networks-in-Persian-language

This is an NLP project, where I am attempting to detect sarcasm in social networks in Persian language.

Overview: This project aims to detect sarcasm in social networks using the Persian language, utilizing cutting-edge techniques such as transformers and the ParsBERT model. Key features: - Detecting sarcasm in Persian language social networks - Utilizing cutting-edge NLP techniques - ParsBERT model for efficient sarcasm detection Tech stack: Python, Transformers, ParsBERT Why it matters: This project aims to develop an accurate and efficient solution for detecting sarcasm within various social networks, leveraging the power of natural language processing (NLP).
TypeScript 5 stars

jbogocz/awesome-automated-ai

A curated list of 250+ tools for Automated AI/ML - from hyperparameter optimization to autonomous AI agents.

### Overview: This curated list offers a comprehensive collection of tools and resources for Automated AI/ML, covering everything from hyperparameter optimization and neural architecture search to autonomous AI agents and automated monitoring. The repository is updated weekly, reflecting the rapid evolution of the field, including advancements in foundation models, autonomous AI, and prompt engineering. ### Key features: - **General-Purpose AutoML**: 22 tools - **Neural Architecture Search**: 6 tools - **Hyperparameter Optimization**: 11 tools - **Automated Feature Engineering**: 9 tools - **Tabular Foundation Models**: 8 tools - **Automl Benchmarks**: 7 tools ### Tech stack: TypeScript ### Why it matters: This list serves as a valuable resource for anyone looking to automate their machine learning pipelines, from ML engineers to researchers and product builders. It covers both the classical tools and the emerging trends in the field, ensuring that the latest advancements are included.
Jupyter Notebook 5 stars

julie-berlin/aie-cohort-7

AI Engineering Bootcamp

Overview: This repository is designed to provide comprehensive training on AI Engineering, focusing on the skills required to build, deploy, operate, and improve Large Language Model (LLM) applications in production environments. Key features: - Prototyping and productionizing LLM application prototypes - Understanding Prompt Engineering, Retrieval Augmented Generation (RAG), Agents, and Fine-Tuning - Productionizing includes evaluating RAG and Agent Applications, improving Search and Retrieval Pipelines, monitoring Production KPIs, setting up Inference Servers, and building LLM Applications with Scalable, Production-Grade Components Tech stack: Jupyter Notebook Why it matters: This bootcamp is essential for data science and engineering teams aiming to successfully build, deploy, operate, and improve LLM applications in production environments, ensuring they are able to build, deploy, monitor, and scale complex LLM applications in production environments. Repository: julie-berlin/aie-cohort-7 GitHub stars: 5 GitHub description: AI Engineering Bootcamp
Python 5 stars

kenza-ily/diagnose_treat_bias_llm

NAACL 2025 | How Can We Diagnose and Treat Bias in Large Language Models for Clinical Decision-Making?

### Overview This repository contains the **Counterfactual Patient Variations (CPV)** framework, a method for detecting and quantifying gender and ethnicity biases in Large Language Models (LLMs) applied to clinical decision-making. CPV constructs counterfactual patient vignettes from medical QA datasets by varying demographic attributes (gender, ethnicity) while holding the clinical content fixed. Bias is measured by comparing LLM outputs across these variations using a suite of lexical and semantic metrics. ### Key Features - **Counterfactual Patient Variations (CPV)**: Constructs counterfactual patient vignettes from medical QA datasets by varying demographic attributes while holding the clinical content fixed. - **Gender and Ethnicity Bias Detection**: Measures bias by comparing LLM outputs across demographic variations. - **Public Datasets**: Three public datasets available on HuggingFace, each built from different medical QA benchmarks: MedQA CPV, MedMCQA CPV, and PubMedQA CPV. ### Tech Stack - Python ### Why It Matters This framework is essential for ensuring fairness and accuracy in clinical decision-making models, particularly in medical applications where patient outcomes can be significantly impacted by demographic factors. By quantifying and detecting bias, it helps in developing more reliable and unbiased AI systems that can improve patient care. ### Installation ```bash git clone https://github.com/kenza-ily/diagnose_treat_bias_llm.git cd diagnose_treat_bias_llm pip install -e . ``` ### Citation ```bibtex @inproceedings{benkirane20
Python 5 stars

khiembk/RECRAFT

Official Implementation of paper "Rethinking Cross-Modal Fine-Tuning: Optimizing the Interaction between Feature Alignment and Target Fitting", which is accepted at AISTATS 2026

Overview: RECRAFT is an algorithm designed to optimize the interaction between feature alignment and target fitting for cross-modal fine-tuning, achieving state-of-the-art performance on NAS-Bench-360 and PDEBench benchmarks. Key features: - Feature Alignment (FA) + Feature-Label Distortion (FLD) - Target Predictor Optimization Tech stack: Python Why it matters: RECRAFT provides a principled approach to cross-modal fine-tuning that minimizes the semantic gap by optimizing the interaction between feature alignment and target fitting, making it highly effective for adapting pre-trained models to unseen feature modalities. Repository: khiembk/RECRAFT GitHub stars: 5 GitHub description: Official Implementation of paper "Rethinking Cross-Modal Fine-Tuning: Optimizing the Interaction between Feature Alignment and Target Fitting", accepted at AISTATS 2026.
Python 5 stars

kimdanny/event-driven-model-serving

Unified API of Apache Kafka and Google PubSub

### Overview: Unified API for Apache Kafka and Google PubSub, enabling efficient data handling and model serving. ### Key Features: - **Unified API**: Simplifies integration of Kafka and PubSub for event-driven model serving. - **Cloud Configuration**: Configures Kafka and PubSub for seamless integration. - **Model Checkpoints**: Stores model checkpoints for model server. - **Queue Simulation**: Simulates Redis queue for model server. - **Database Simulation**: Simulates a simple database for application 2. ### Tech Stack: Python ### Why it Matters: This project provides a unified and efficient solution for integrating Kafka and PubSub, making it easier to manage and scale event-driven model serving systems.
Jupyter Notebook 5 stars

klay-liu/Financial-Intent-Understanding-with-LLMs

🎯 Fine-tuning LLMs using LlamaFactory for financial intent understanding | Evaluating open-source models on OpenFinData benchmark | Full implementation with multiple models (Qwen2.5/ChatGLM3/Baichuan2/Llama3)

### Overview: 本项目展示了如何使用LlamaFactory微调大语言模型以解决金融领域的特定任务。我们选取了开源金融评测数据集[OpenFinData](https://github.com/open-compass/OpenFinData)中的金融意图理解任务作为示例，评估了主流开源大模型在该任务上的能力，并探索了LoRA微调对模型性能的提升效果。 ### Key features: - 使用LlamaFactory微调大语言模型 - 评测主流开源模型在金融意图理解任务上的能力 - 探索LoRA微调对模型性能的提升效果 - 提供一个完整的LLM微调实践案例 - 评估主流开源模型在金融意图理解任务上的能力 - 探索LoRA微调对模型性能的提升效果 - 为金融领域LLM应用提供参考 ### Tech stack: Python 3.8+ ### Why it matters: 通过本项目，开发者可以了解如何利用开源工具和模型进行金融领域的微调，从而提升模型在实际应用中的表现。此外，通过对比不同模型的性能，开发者可以更好地选择适合自己需求的模型，提高应用效果。
Jupyter Notebook 5 stars

knetic0/turkish-store-reviews-sentiment-analyses

🔥 Sentiment analysis on Turkish store reviews using LSTM, BiLSTM, and GRU models. Interactive UI built with Gradio.

### Overview: This project analyzes sentiment (positive, neutral, negative) on Turkish store reviews using LSTM, BiLSTM, and GRU models. The user interface is built with Gradio for interactive analysis. ### Key features: - Sentiment analysis on Turkish store reviews - Multiple models: LSTM, Bidirectional LSTM, GRU - Gradio for interactive UI - 440,679 training data, 48,965 test data ### Tech stack: Jupyter Notebook ### Why it matters: This project provides a comprehensive solution for sentiment analysis of Turkish store reviews, utilizing advanced machine learning techniques and an interactive user interface for easy analysis.
Python 5 stars

l11x0m7/LMPresent

Including pre-trained language models for fine-tuning on other NLP tasks

Overview: This repository includes pre-trained language models (BERT, ELMo, and GPT) for fine-tuning on various NLP tasks, such as the RACE dataset, WikiQA, SQuAD, and ROCStories. Key features: - Pre-trained models for fine-tuning on the RACE dataset - Fine-tuning experiments on WikiQA and SQuAD - Detailed results and analysis for BERT, GPT, and ELMo models - Experiments on the ROCStories dataset Tech stack: Python Why it matters: This repository provides a comprehensive set of pre-trained models and fine-tuning experiments, making it a valuable resource for extending pre-trained models to other NLP tasks. Repository: l11x0m7/LMPresent GitHub stars: 5 GitHub description: Including pre-trained language models for fine-tuning on other NLP tasks
Python 5 stars

leonvanbokhorst/friction-reasoning

A friction reasoning conversational data set generator and reasoning model fine-tuning

### Overview: Pushback-AI is a comprehensive toolkit designed for generating and fine-tuning datasets for disagreement-driven AI. It includes synthetic dialogue generation, persona orchestration, dataset packaging, fine-tuning using LoRA and DeepSeek-R1, GGUF export, and deployment via Hugging Face and Ollama. ### Key Features: - **Disagreement-first dataset synthesis**: Six contrasting agent personas (skeptical, memory-rich, mechanical, contrarian, humble, synthesizer) to create productive disagreement. - **Prompt templating system**: Enforces gestures, hesitation, and persona tone. - **LoRA fine-tuning pipeline**: Targeting DeepSeek-R1 distilled models using Unsloth. - **Deployment tooling**: GGUF export for Hugging Face/Ollama publication. ### Tech Stack: Python ### Why it Matters: The end-to-end workflow (generation → training → export) is tested on **Windows 11 WSL2** with Python 3.10–3.12. macOS builds may require alternative GPU backends (e.g., replacing `bitsandbytes`). ### Repository: - **GitHub**: [leonvanbokhorst/friction-reasoning](https://github.com/leonvanbokhorst/friction-reasoning) - **GitHub Stars**: 5 - **GitHub Description**: A friction reasoning conversational data set generator and reasoning model fine-tuning ### Documentation Hub: - **Phase Docs**: - 01: Synthetic dataset pipeline - 02: Prompt
HTML 5 stars

m1ns09/Llama

🌐 Run GGUF models directly in your web browser using JavaScript and WebAssembly for a seamless and flexible AI experience.

Overview: Llama is a web application that allows you to chat with a friendly AI character directly in your browser. It utilizes cutting-edge artificial intelligence for seamless and flexible conversations. Key features: - Interactive Conversations - User-Friendly Interface - Customizable Settings - Fast Response Times Tech stack: JavaScript, WebAssembly Why it matters: Llama provides a unique and engaging way to interact with AI, offering a natural and flexible chat experience directly in your browser.
Python 5 stars

marcelo-earth/light-tuning

🤖 CLI to fine-tune GPT-3.5 Turbo models rapidly. 💬

Overview: LightTuning is a CLI tool designed to rapidly fine-tune GPT-3.5 Turbo models, making it easier to scale and enhance models from small conversations to high-performance models that outperform GPT-4. Key features: - Generate dataset - Upload dataset - Fine-tune model - All Tech stack: Python Why it matters: LightTuning simplifies the process of fine-tuning GPT-3.5 Turbo models, making it accessible and efficient for developers and enthusiasts alike. Repository: marcelo-earth/light-tuning Primary language: Python GitHub stars: 5 GitHub description: 🤖 CLI to fine-tune GPT-3.5 Turbo models rapidly. 💬 README excerpt: 🚀 Concept: LightTuning is a quick way to fine-tune GPT-3.5 Turbo models. 🧑‍💻 Usage: Install the required packages, run the script, and choose actions like generating a dataset, uploading a dataset, or fine-tuning the model. 📃 License: Distributed under the MIT License.
Jupyter Notebook 5 stars

merishnaSuwal/nep-off-langdetect

Offensive language detection and sentiment analysis on Nepali text

Overview: This repository contains the code and datasets for the Nepali Offensive Language Detection and Sentiment Analysis project, focusing on the identification of offensive entities and sentiment classification in Nepali text. Key features: - In-depth exploration of linguistic challenges in offensive language detection for a multilingual context. - Evaluation of pre-trained BERT architectures for Nepali, with a particular emphasis on Aspect term extraction using Named Entity Recognition (NER) and comprehensive analysis of sentiment classification in detecting Offensive Nepali language. Tech stack: Python 3.x, Jupyter Notebook, PyTorch, Transformers, scikit-learn, pandas, numpy, matplotlib, seaborn, and any other dependencies mentioned in the notebooks. Why it matters: This project aims to enhance content moderation tools and create safer online spaces for the Nepali-speaking community by addressing the limitations of pre-trained models in multilingual contexts.
Jupyter Notebook 5 stars

morteza89/NLP-LLM

Overview: This repository explores cutting-edge methods in Natural Language Processing (NLP) and Large Language Models (LLMs), focusing on techniques such as Combined Topic Modeling (CTM), fine-tuning Llama 2 for sentiment analysis, and constructing multimodal RAG pipelines. Each section delves into specific aspects of NLP-LLM, from topic discovery to sentence sentiment classification and translation, showcasing the practical applications of advanced NLP techniques. Key features: - Fine-tune Llama 2 for sentiment analysis - Construct multimodal RAG pipelines - Sentence sentiment classification with DistilBERT - Sentence-to-sentence translation - Design of RAG using Llama-2, Langchain, and Chromadb Tech stack: Jupyter Notebook Why it matters: These projects highlight the practical applications of cutting-edge NLP techniques and demonstrate the synergistic integration of various tools and frameworks to solve complex linguistic challenges, pushing the boundaries of what can be achieved in the field of natural language understanding and generation. Repository: morteza89/NLP-LLM GitHub stars: 5 GitHub description: No description provided. README excerpt: # NLP-LLM ### This repository is an ongoing project aimed at exploring state-of-the-art (SOTA) methods in NLP, and will continue to evolve with more practical algorithms and approaches... For Natural Language Processing (NLP) and Large Language Models (LLMs), the exploration and implementation of innovative methodologies like Combined Topic Modeling (CTM), fine-tuning Llama 2 for sentiment analysis, constructing multimodal
Python 5 stars

nanofatdog/train-studio

Train Studio - SFT LoRA Web UI (Gradio): hardware detect, GPU select, config, train/resume, merge tool

Overview: Train Studio is a web-based user interface designed for training Low-Rank Fusion (LoRA) models using SFT (Scaling Fusion Training) on GPU hardware. It allows users to select GPU, model, dataset, and configuration parameters directly through the browser, providing a seamless experience akin to unsloth.ai. Key features: - Hardware detection and selection - Model and dataset configuration - Training configuration management - Real-time training status and log refresh - Automatic LoRA merge and full model generation Tech stack: Python, Gradio, Hugging Face Transformers, PyTorch, CUDA, GPU, GPU memory management, and GPU-specific libraries like nvidia-smi and /proc. Why it matters: Train Studio simplifies the process of training LoRA models, making it accessible to developers and researchers without requiring deep expertise in GPU management or complex training scripts. It streamlines the workflow by automating common tasks and providing a user-friendly interface, enhancing productivity and reducing the learning curve for new users.
TypeScript 5 stars

natekali/dataforge

The Ultimate Fine-Tuning Dataset Builder: a human-centered, AI-powered platform for LLM dataset creation

### Overview: DataForge is a comprehensive and user-friendly platform designed specifically for fine-tuning Large Language Models (LLMs) datasets. It seamlessly integrates with your existing data sources and offers a streamlined workflow for generating, editing, and exporting datasets tailored to modern LLMs. ### Key Features: - **Import**: Supports a wide variety of data formats including JSONL, JSON, CSV, Parquet, Excel, PDF, DOCX, Markdown, and plain text. It also imports datasets directly from Hugging Face Hub. - **Dataset Types**: Manages four types of datasets: SFT chat conversations, DPO preference pairs, KTO unpaired feedback, and RL prompts. - **Reasoning Traces and Tool Calls**: Keeps these data fields separate to ensure correct rendering for each target model. - **Edit**: Features a virtualized grid that handles large datasets without performance issues, including role switching, reasoning blocks, tool call cards, and per-message loss masking. - **Quality Checks**: Includes 17 issue checks for empty messages, broken tool calls, PII, encoding damage, and context overflow. - **Generate**: Offers synthetic data generation, Q&A from documents, and on-policy DPO pair building. - **Export**: Provides various export options for compatibility with different LLMs, including JSONL, Axolotl 0.17, TRL 1.5, LLaMA-Factory 0.9.5, MS-SWIFT 4.3, Unsloth, and OpenAI fine-tuning. - **Model Registry**: Tracks
Python 5 stars

naufalahnaf17/pytorch-cv-project

several computer vision projects in learning to create deep learning models, especially in the field of computer vision, starting from classification, object detection and others

### Overview: These projects showcase various computer vision tasks using PyTorch, including image classification, object detection, and text detection. The author fine-tunes pre-trained models and exports them to ONNX format for easy deployment. ### Key Features: - **Image Classification**: Fine-tuned ResNet-50 and Google VIT-Base-Patch16-224 for cats and dogs, and plant species respectively. - **Object Detection**: Fine-tuned Faster-RCNN and YOLO-V10 for text detection and facemask detection, respectively. - **Text Detection**: Fine-tuned Faster-RCNN for manga text detection. ### Tech Stack: Python ### Why it Matters: These projects demonstrate practical applications of PyTorch in computer vision tasks, showcasing the effectiveness of fine-tuning pre-trained models and the importance of model export to ONNX format for deployment.