Lightning AI

From the PyTorch Lightning creatorsThe AI cloud PyTorch developers loveLightning is the AI cloud for developers and AI teams that makes it easy to build and deploy lightning fast models.Our experts and AI copilots help at every step.Get started freeRequest demoAI StudioExploreCollaborative GPU cloud workspace where AI helps debug, train, inference.AI notebooksExplorePersistent GPU notebooks where AI helps you code and analyze datasets.GPU clustersExploreManaged training, inference clusters. SLURM, K8s or our multi-cloud LEC.InferenceExploreUse pay-per-token APIs, serve custom models or let us do it all for you.Trusted by 340,000+ devs and AI teamsStart your next AI project from a templateAllRLAgentsChatbotsAI appsInferenceTrainingScienceView allRun Clawdbot instantly on the cloud. No Mac minis, no local installs, no terminal access. Secure, isolated cloud environment with one-click launch and browser-based setup.Clawdbot in the cloud, zero setup (openclaw)997 clones9.05K Viewsqrl-qai is the quantum analogue of OpenAI's gym python frameworkqrl-qai-0.2.0 playground6 clones1.03K ViewsBuild a reasoning LLM with reinforcement finetuning of Qwen 3 using GRPO & Unsloth. Complete local tutorial for math/reasoning tasks with proximity-based rewards & HuggingFace datasets. Includes training & inference notebooks. 100% local setup.Build a reasoning LLM from scratch using GRPO89 clones19.15K ViewsSheepRL Workshop: add Super Mario Bros environment and train an agent on it.SheepRL: How to integrate Super Mario Bros enviroment31 clones1.29K ViewsIn this project, I set out to merge three powerful methodologies—LSTM, XGBoost, and Q-Learning—to decode the chaotic rhythm of the stock market and optimize decision-making under uncertainty.Uncovering Patterns in Stock Market Data: Predictive Analytics and Actionable Recommendations18 clones855 ViewsRun Clawdbot instantly on the cloud. No Mac minis, no local installs, no terminal access. Secure, isolated cloud environment with one-click launch and browser-based setup.Clawdbot in the cloud, zero setup (openclaw)997 clones9.05K ViewsLeverage the power of Crew AI to build an intelligent multi agent system. Where each agent has a dedicated task & role. Agents can also interact & delegate task amongst each other. The system we build here scrapes web to automatically write an engaging blog post for you.Build a crew of AI agents371 clones6.58K ViewsDeploy AI Agent with Tool use using OpenAI compatible API format. Deploy AI Agent with Tool Use107 clones3.29K ViewsDiscover the future of automated flight booking with our multi-agent AI system. This solution uses CrewAI, Browserbase, and Ollama to autonomously find and summarize the best roundtrip flights—all while running completely locally to protect your privacy.Build a multi-agent flight booking crew using DeepSeek-R188 clones7.17K ViewsLearn to build an Agentic RAG app powered by Qwen 3 that searches a vector DB and falls back to web search if needed.Agentic RAG powered by Qwen 393 clones5.43K ViewsThis Studio uses SheepRL to train a PPO agent on the cartpole environment.Train RL Agents with SheepRL17 clones1.45K ViewsAnalyze and research financials of public listed companies in minutes using an AI research agent. Stock Research Agent with LitServe103 clones18.96K ViewsChat with you docs using RAG, featuring newly released Llama-3 & Phi-3. See for yourself which model performs better using an interactive Chat application created using Streamlit.Compare Llama-3 and Phi-3 using RAG262 clones16.95K ViewsMulti-Document Agentic RAG for Quantum Computing28 clones3.17K ViewsLeverage LLMs to share engaging commentary on AI content on LinkedIn and Twitter.Social Media Influencer AI-powered bot120 clones2.64K ViewsThis interactive chatbot can help you find information about medicines from patient reviews. You can ask questions about treatment, drug interactions, side effects etc. The chatbot uses a combination of OpenAI's text embeddings and Pinecone's vector store to find relevant informationDrug Review AI Chat bot9 clones1.14K ViewsDiscord bot that answers unanswered questions using GPT-4, web search, and memory, built with Agno AGI and Supabase for community support.Discord Bot with Agno3 clones883 ViewsDiscover a fresh approach to interact with your documents through MetaAI's Llama-3, served locally using Ollama. Query your knowledge base effortlessly to pinpoint exactly what you need using a natural language interface.RAG using Llama 3.1 by Meta AI1.15K clones37.68K ViewsCreate personalized travel plans easily with the AI Travel Planner app. Just enter your destination, travel dates, and interests to get custom recommendations generated by automated AI Agents.AI Travel Planner (CrewAI+Ollama)99 clones2.45K ViewsThe template allows a user to chat with SQL Data in Natural Language via a Streamlit AppChat with your SQL Database38 clones1.88K ViewsBase (minimalist) implementation of https://github.com/comfyanonymous/ComfyUI as a Lightning AI Studio. Just has an upgraded version and the extension manager. This studio is designed for rapid start-up and development of new projects in ComfyUIClean ComfyUI Template v0.3.15 20250221609 clones2.60K ViewsWe'll use CLIP, a model that understands images based on language, and host a streamlit app.Host an AI web app with OpenAI's CLIP model8.85K clones24.03K ViewsDeploy a real-time voice transcription web app using Open AI's Whisper model on cloud GPUs.Deploy a real-time voice transcription app121 clones2.15K Viewszyphra/zonos v0.1.0 for text-to-speech and voice synthesis, published by mathematicalmichael with dependencies pre-installedzyphra/zonos:v0.1.049 clones1.66K ViewsPhi-3-vision is a lightweight, state-of-the-art 4.2 billion parameter multimodal model with language and vision capabilities, available with a 128k context length.Deploy and chat with Phi-3-vision-128k-instruct189 clones4.22K ViewsExplore & learn how to create a self-hosted document summarization and chat application using Llama-3 and Ollama, complete with a demo and step-by-step guide to launch your very own RAG app.Document summarization & chat RAG application92 clones3.54K ViewsRun R Studio Server in a Lightning StudioRun R Studio in the Cloud47 clones979 ViewsGet up and running quickly with Llama 3 8B with a fully customizable UI in pure PythonReflex Chat App - Llama 3113 clones2.18K ViewsQuery your knowledge base effortlessly without compromising your privacy. Learn how to build a self-hosted document chat RAG application using Cohere's powerful Command R model and a locally served vector database, Qdrant.Self hosted RAG app using Cohere's ⌘R199 clones5.78K ViewsLearn to set up a fully local MCP client—no cloud, no data leaks. Step-by-step guide to offline installs, encryption, and airtight security best practices.Build a Private & Secure MCP Client (100% Local)140 clones10.90K ViewsServe Image Classification model using FastAPI for free. Learn to host Python web apps with FastAPI and Lightning Studio ⚡️Deploy Machine Learning API with FastAPI for free112 clones6.10K ViewsBenchmark vLLM - 2x faster LLM Inference with AWQ (4-bit quantization) for Mistral 7B using vLLM. Learn how to serve open LLMs with OpenAI API protocol and run quantized LLMs in production.Optimized LLM inference API for Mistral 7B using vLLM124 clones17.06K ViewsRun Mistral AI's Mixtral LLM and chat with it in a Streamlit UIRun Ollama on a Cloud GPU Lightning Studio676 clones7.64K ViewsDeploy ANY Hugging Face model instantly with LitServe. This Studio shows the simple steps to get a Private API endpoint for these models with full code access and control.Deploy a Private API for any Hugging Face model332 clones42.24K ViewsDiscover the future of conversational AI with DeepSeek-R1, a state-of-the-art reasoning model. Powered by LitServe and OpenAI-compatible APIs, it offers real-time engagement, quick deployment, and unmatched scalability. Chat with DeepSeek-R1, an Advanced AI Reasoning Model 🤖275 clones19.26K ViewsDeploy a private instance of Stability AI's Stable diffusion 2 on GPUs.Deploy a private API for Stable diffusion 2223 clones4.05K ViewsDeploy a private instance of Llama 3 (8B) on a private, self-managed API with full code access and control.Deploy a private Llama 3 (8B) API282 clones5.84K ViewsTurn written words into speech that sounds like a particular voice. This Studio deploys a voice mimic API. Uses the Coqui XTTS V2 model.Deploy a voice clone API - Coqui XTTS V2 model591 clones8.05K ViewsThis studio shows how to deploy a RAG system on text documents using OpenAI and LitServe.Deploy an OpenAI RAG Server105 clones3.90K ViewsDeploy a voice clone API with F5 text to speech, LitServe and a Lightning StudioDeploy voice clone API (F5 TTS)248 clones6.11K ViewsRF-DETR is a groundbreaking, real-time, transformer-based next-generation model developed by Roboflow, pushing the boundaries of real-time performance. This studio provides a comprehensive guide to deploying the state-of-the-art RF-DETR model using LitServe. Deploy RF-DETR: A SOTA Real-Time Object Detection Model using LitServe48 clones10.52K ViewsKokoro-82M is a high-performing, compact TTS model released under the Apache license. It supports a variety of languages, as well as multiple voices, and ranks highly despite limited training data. This studio shows how to create a self-hosted, private API that deploys the model with LitServe.Deploy Kokoro TTS model60 clones5.68K ViewsVideoLLaMA 3 excels at understanding images and videos, employing advanced architecture for superior visual data processing and complex reasoning across diverse settings. This studio demonstrates a private, self-hosted API for the VideoLLaMA 3 multimodal model, leveraging LitServe.Deploy VideoLLaMA 3 Multimodal model12 clones1.18K ViewsThis studio creates a image segmentation API using Litserve and Segment Anything 2, which can extract the subject from a photo. Deploy a image segmentation API with Meta's SAM267 clones2.55K ViewsDeploy a multi-modal LLM that can process images and text with PixtralDeploy a multi-modal LLM with Pixtral89 clones4.49K ViewsLearn to deploy Retrieval-Augmented Generation (RAG) applications using LitServe for scalable, serverless access with multi-GPU support. Simplify your AI deployments.Deploy a private Llama 3.1 RAG API168 clones6.86K ViewsInstantly remove image backgrounds with our powerful API, seamlessly deployed using LitServe. Fast, reliable, and easy to integrate—perfect for any application needing quick background removal.Deploy Background Removal API with LitServe21 clones1.39K ViewsLearn how to deploy a XGBoost model as a private API with LitServe. This tutorial covers real-world use cases like classification, regression, and anomaly detection.Deploy XGBoost with LitServe36 clones2.27K ViewsDeploy a voice clone API that generates audio spoken by the target voiceDeploy a voice clone API (SVTT2)55 clones2.61K ViewsDeploy a private instance of Open AI's Whisper model on GPUs.Deploy a private API for Open AI's Whisper model179 clones4.46K ViewsDeploy a BERT hugging face model with LitServe. This Studio creates a private, self-managed API with full code access and control.Deploy a Hugging Face BERT model67 clones3.13K ViewsFinetune the optimal model by trying different hyperparameter combinations using grid search or random search.Run a hyperparameter sweep2.67K clones8.41K ViewsIn this tutorial, you'll learn to train a time series forecasting model using PyTorch Lightning with historical stock price data. We'll leverage a pre-trained sequence model from PyTorch's library, guiding you through dataset setup, model architecture, and training process.Time-Series Forecasting with PyTorch Lightning127 clones7.14K ViewsThis studio shows how to finetune and serve Meta AI's Llama 3.2 language models.Finetune and Serve Llama 3.2 1B and 3B388 clones7.51K ViewsPretrain a large language model (LLM) with PyTorch Lightning and LitData. This Studio is used in the README for PyTorch Lightning.Pretrain an LLM with PyTorch Lightning47 clones3.23K ViewsTrain a diffusion model from scratch to generate realistic images. This Studio is used in the README for PyTorch Lightning.Train a diffusion model with PyTorch Lightning 137 clones9.64K ViewsFinetune a simple audio generation model on your own music collection with PyTorch Lightning. This Studio is used in the README for PyTorch Lightning.Finetune a personal AI music generator173 clones5.99K ViewsGuide for getting started with LitGPT - Load, pretrain, finetune, deploy 20+ LLMs on your own data. Uses state-of-the-art techniques: flash attention, FSDP, 4-bit, LoRA, and more.LitGPT quick start167 clones51.45K ViewsDiscover how UnslothAI dramatically accelerates LLM finetuning and reduces memory usage. Get started with our hands-on guide to optimize your models for faster inference and better performance.UnslothAI: Accelerate LLM finetuning!172 clones2.78K ViewsGemma is Google’s latest open-weight LLM. This Studio explains shows you how to use Gemma through Lit-GPT and explains some of the unique design choices of Gemma compared to other LLMs.Understanding, Using, and Finetuning Gemma465 clones24.34K ViewsLearn how to take a pretrained model and continue pretraining it on a new dataset of your choice.Continued Pretraining with TinyLlama 1.1B191 clones8.47K ViewsIn this quest, we'll finetune Open AI's CLIP model to a small dataset. We'll start on a CPU Studio and use a GPU to 36x the speed. We'll also monitor the finetuning with Tensorboard and share the public link with colleagues.Finetune a pretrained model4.33K clones15.52K ViewsRun R Studio Server in a Lightning StudioRun R Studio in the Cloud47 clones979 ViewsUnleashing the Potential of MONAI Framework for Advanced 3D Lung Tumour Segmentation ResearchEmpowering 3D Lung Tumour Segmentation with MONAI49 clones1.73K ViewsRun a Physics Simulator in Studios27 clones720 ViewsWe introduce MDLM, a Masked discrete Diffusion Language Model that features a novel (SUBS)titution based parameterization which simplifies the absorbing state diffusion loss to a mixture of classical masked language modeling losses. In doing so, we achieve SOTA perplexity numbers on LM1B and OpenWebSimple and Effective Masked Diffusion Language Models7 clones1.11K ViewsHands-on machine learning with fNIRS using the BenchNIRS Python frameworkBenchNIRS Workshop0 clones594 ViewsThe DEEP-EM TOOLBOX supports the application and adaptation of deep learning (DL) solutions in Electron Microscopy (EM) labs, bridging the gap between DL experts and EM researchers. By introducing standardized workflows and plug-and-play use cases, the toolbox fosters interdisciplinary collaborationDEEP-EM TOOLBOX: Segmentation of Cellular Structures18 clones758 ViewsTraditional clouds weren’t built for AI.Our AI cloud is.It takes more than a model and GPUs to ship AI. Lightning gives you specialized tools, reliable GPU infrastructure, and expertise to ship high-performance AI.Featured toolsTeamsUsecasesAI StudioCode and build with AI help on a persistent cloud workspace that feels local.AI notebooksCode and analyze data with AI help with persistent, collaborative notebooks.Model APIsCall AI models through a pay-per-token API. 30M free tok/month.Batch jobsRun batch work on 1000s of GPUs with fractional, pay-as-you-go pricing.InferenceServe custom models with full control or let us do it all for you.GPUsOn-demand VMs, reserved or ephemeral clusters with K8s, Vast, infiniband, etc.Explore all toolsOne account. GPUs from everywhere.AWSGCPLightningLambdaNebiusNScaleVoltageGet better ratesCPU1 GPU2 GPU4 GPU8 GPUClustersNameGPUsVRAM*Cost/GPU/hr*Cost/GPU/hr* (interruptible)Free hours monthly**T4116 GB$0.19$0.51﹘$0.6175L4124 GB$0.48$0.47﹘$0.5731L40S148 GB$2.89$1.47﹘$1.765RTXP 6000196 GB$2.89$1.43﹘$1.722A100140 GB$1.29 - 10A100180 GB$2.71$3.06﹘$3.675H100180 GB$1.99 - 5H2001141 GB$3.50 - 3*VRAM and price per GPU. Billed by the second.**Free GPU hrs with 15 monthly free credits.Loved by developers. Trusted by IT.Give developers the freedom to build - with the guardrails IT demands.Hardened over 5+ years in the enterprise.SOC2HIPAAGDPRRequest demoSet budgets by team or projectReal-time cost trackingAutosleep idle computeSSO + role-based accessFine-grained data accessAudit logsSOC2 & HIPAA compliancePrivate cloud & VPC supportEncryption at rest & in transitBring your own cloudRuns on your stack (K8s, Slurm)Elastic multi-cloud portabilityBuild something real today.Get started freeRequest demo --- Get 80 free GPU hours monthly.Then pay-as-you-go.Start on the free tier with 80* free GPU hours. Pay as you go for more hours. Upgrade for enterprise features like bring your own cloud, use your AWS/GCP credits, SOC2, HIPAA, and 24/7 support.Run 1 Studio free 24/7**. No credit card. No commitments.Get started free*Prices vary**Free Studios run 24/7 but require restart every 4 hours. No restrictions in Pro or higher.CPU1 GPU2 GPU4 GPU8 GPUClustersNameGPUsVRAM*Cost/GPU/hr*Cost/GPU/hr* (interruptible)Free hours monthly**T4116 GB$0.19$0.51﹘$0.6175L4124 GB$0.48$0.47﹘$0.5731L40S148 GB$2.89$1.47﹘$1.765RTXP 6000196 GB$2.89$1.43﹘$1.722A100140 GB$1.29 - 10A100180 GB$2.71$3.06﹘$3.675H100180 GB$1.99 - 5H2001141 GB$3.50 - 3*VRAM and price per GPU. Billed by the second.**Free GPU hrs with 15 monthly free credits.Access the world's best clouds from one place.AWSGCPLightningLambdaNebiusNScaleVoltageLightning on your private cloud.Use AWS/GCP commits, keep data private in your VPC. Run a secure, customized AI platform, without managing infrastructure.Request demoUse cloud creditsMaximize AWS/GCP commitments.Fully customizableBuild your dream AI platform - zero infra management.SOC2 · HIPAAEnterprise-grade security for your data.Start Small, Scale BigFrom solo projects to enterprise AI, choose a plan that grows with you.Academic pricingFree$0Students, researchers, hobbyistsChoose Free15 monthly creditsKey features1 free active Studio (4 hour restarts)32 core CPU StudiosSingle GPUs (L40S, A100, H100, H200...)80% off on interruptible (spot)Up to 2 concurrent GPUsPersistent storage (50 GB limit)15 req/min to our model APIs120,000 tok/min to our model APIsConnect any local IDE or SSHUnlimited background executionMultiplayer live collaborationUse private and public modelsAccess optimized StudiosAutomate with our SDKPay as you goProMonthlyAnnual$50-60%$20per month(billed annually)Developers, researchers, scientistsChoose Pro240 annual creditsEverything in Free, plus:1 active Studio (no restarts)64 core CPUMulti-GPU (T4, L4, L40S)Multi-node training (6 GPUs max)Up to 6 concurrent GPUsPersistent storage (200 GB limit)20 req/min to our model APIs120,000 tok/min to our model APIsConnect public, private S3 bucketsDistributed data prep (up to 4 machines)Reserve machines for jobsPay as you go for more creditsTeamsMonthlyAnnual$140-15%$119/user, per month(billed annually)Research labs, small teamsChoose Teams600 annual creditsEverything in Pro, plus:1 active Studio (no restarts)96 core CPUFull-node A100, H100, H200Multi-node training (12 GPUs max)Up to 12 concurrent GPUsPersistent storage (2 TB limit)30 req/min to our model APIs150,000 tok/min to our model APIsUse via AWS marketplaceReal-time cost controlsSpending limitsEnterpriseCustomEnterprise-grade AIRequest a quoteCustom creditsEverything in Teams, plus:Use your cloud credits (AWS, GCP)Priority GPU Access (Lightning Cloud)Full-node B200sUnlimited multi-node trainingUnlimited concurrent GPUsDeploy in your own VPCCustom resource taggingCustom rate limits for our model APIsEnterprise AI Hub add-on99.95% uptime SLASOC 2 (type 2) complianceSAML/SSOBring your own imagesDedicated Slack support channelDedicated machine learning engineerRole-based access controlsYou have to try it to believe it.350,000+ builders love Lightning AIPrototype 6× faster"Instead of taking 36 hours to know if this model works, we knew by the end of the day."Sam, ResearcherMIT Sustainable Design LabExperiment at lightning speed"We validated our hypothesis in just 4 weeks for under $10k. It would've taken us at least 70% longer if we did not have the Lightning platform!"Malcolm, co-founderOctAIDebug together"When it came time for our experts to collaborate on common experiments we ran into the “it works on my machine” problem. With Lightning, we saved hours per week."Robin Cole, ResearcherEarthDaily AnalyticsChange GPUs in seconds"The experience of changing from CPU to GPU was mindblowing. It’s extremely quick and I didn’t even need to think about copying or mounting my files."Irina, ML EngineerLightning UserNo environment setup"The environment management is so much better with Lightning AI; like an environment already provisioned for a workspace is a really good model."Jacqueline, Sr. ML EngineerAeiraCompare featuresStartup and academic pricing available. Annual pricing available. Get in touch.ComputeStudio runtimeDataCollaborationAI workflowsSecuritySupportCurrentComputeFree active CPU StudiosFree active CPU Studio session limitMax inactive StudiosFree creditsAdditional CPU/GPU creditsMax CPUs per StudioMax GPUs per StudioT4, L4, L40S session limitA100, H100, H200 session limitMax concurrent GPUsCloud providersInterruptible instancesChange Studio between CPU and GPUsReserve machines for jobsUse without an AWS or GCP accountBasic cloud cost managementOrganization level cloud cost managementUse your own AWS or GCP accountSet up a custom machineStudio runtimeReady-to-code (no CUDA setup, etc...)Setup once, reuse any timeCode from the browserConnect local IDE (VSCode, PyCharm, ...)Notebooks on the browserSudo accessTerminal accessConnect via SSHBackground execution limitMonitor system metrics real-timeStudio auto sleepCustomize auto sleep behaviorAuto switch to CPU when inactiveBuild templates with Base StudiosAdvanced configuration for Base StudiosCustom resource taggingBring your own imagesConnect internal pip/conda mirrorsDataVisualize/explore datasetsUpload dataConnect S3, GCS and R2 bucketsCreate cloud-agnostic foldersHigh-performance storage (EFS/Filestore)Free storagePersistent storage limitAdditional storageCollaborationMultiplayer live collaborationPrivate shared file systemPublish public StudiosPublish private StudiosControl data accessCreate organizationsAuto joinCredits Auto-ReloadTeamspacesSeat priceAI workflowsPublic modelsPrivate modelsCommunity templatesPretrain, finetune modelsMulti-node training (max GPUs)Foundation model auto-babysittingDistributed data prepDistributed data prep (max CPU machines)Host and share AI appsShare public app linksDeploy no-code model endpointsDeploy full control model endpointsAccess Studio pluginsBuild your own QuestsAdd your own pluginsDesign your own ML platformEnterprise AI Hub add-onModel APIs req/minModel APIs tok/minSecuritySecretsEnd-to-end data encryptionSAML/SSOUse your own AWS accountLightning in your VPCKMS EncryptionLightning on premRole-based access controlsSOC2 (type 2) complianceBring your own encryption keysAudit logsBring your own imagesConnect to internal pip/conda mirrorsSupportLearn AI and deep learningPyTorch Lightning supportPlatform supportUptime SLAExpert model performance tuningGet a custom gen AI model trained, deployedAI strategy designCurrentFreeStudents, researchers, hobbyists14 hoursUnlimited15 monthlyPay-as-you-go321Unlimited4 hours2All----Unlimited--1------10 GB50 GBPay-as-you-go-Unlimited$0------15120,000-----------Visit our coursesJoin our 6k+ communityJoin our 6k+ community----CurrentProDevelopers, researchers, scientists1UnlimitedUnlimited40 monthly / 240 annuallyPay-as-you-go644UnlimitedUnlimited6All---Unlimited1-----10 GB200 GBPay-as-you-go-Unlimited$064----20120,000-----------Visit our coursesJoin our 6k+ communityJoin our 6k+ community----CurrentTeamsResearch labs, small teams1UnlimitedUnlimited50 monthly / 600 annuallyPay-as-you-go1928UnlimitedUnlimited12All--Unlimited1-----10 GB2 TBPay-as-you-goUnlimited$140 per user/month1232----30150,000-----------Visit our coursesJoin our 6k+ communityJoin our 6k+ community----CurrentEnterpriseEnterprise-grade AI1UnlimitedUnlimitedCustomBulk pricing1928UnlimitedUnlimitedUnlimitedAllUnlimitedUnlimited10 GBUnlimitedUnlimitedUnlimitedCustomUnlimitedUnlimitedCustomCustomOnsite classes available24/7 expert support24/7 expert support99.95%FAQsWhat do you mean by running 24/7?You can create unlimited Studios. At any one time, you can run one (CPU) Studio for free. If you want to run more at the same time, you must use your free credits or buy more credits.Do you offer a trial?Yes! The free tier is truly free! You can run 1 Studio for free and get 15 free credits per month. 15 free credits gets you ~80 GPU hours per month on interruptible machines. Prices may varyWhy do you need my phone number?To prevent abuse of the platform.How do the 15 free credits per month work?Every month, we give you 15 free Lightning credits. If you don't use them, they expire. Use the credits to run additional Studios or for GPU hours.When do free credits expire?Free credits expire every month. Make sure to use them up!When do purchased Lightning credits expire?After 12 months.How do I use AWS/GCP credits?Upgrade to the enterprise tier to use your AWS/GCP credits.Which cloud providers does Lightning support?Currently AWS and GCP. Azure support is coming soon. --- ABOUT LIGHTNING AILightning empowers everyone to build AI.Our platform provides intuitive open source tools,powerful cloud infrastructure and expertise to help you build AI securely.NVIDIANVIDIA NeMoMicrosoftTorchGeoStability AIStable diffusionPlayground AIText to imageMetafastMRINVIDIABioNeMo - Drug discoveryMicrosoftCameraTraps - Detect wildlifeRunway MLGen 3 familyNVIDIANVIDIA NeMoMicrosoftTorchGeoStability AIStable diffusionPlayground AIText to imageMetafastMRINVIDIABioNeMo - Drug discoveryMicrosoftCameraTraps - Detect wildlifeRunway MLGen 3 familyNVIDIANVIDIA NeMoMicrosoftTorchGeoStability AIStable diffusionPlayground AIText to imageMetafastMRINVIDIABioNeMo - Drug discoveryMicrosoftCameraTraps - Detect wildlifeRunway MLGen 3 familyNVIDIANVIDIA NeMoMicrosoftTorchGeoStability AIStable diffusionPlayground AIText to imageMetafastMRINVIDIABioNeMo - Drug discoveryMicrosoftCameraTraps - Detect wildlifeRunway MLGen 3 familyNVIDIANVIDIA NeMoMicrosoftTorchGeoStability AIStable diffusionPlayground AIText to imageMetafastMRINVIDIABioNeMo - Drug discoveryMicrosoftCameraTraps - Detect wildlifeRunway MLGen 3 familyFrom open source to cloud platformLightning AI is the creator of PyTorch Lightning, the framework for training and finetuning AI models, and Lightning AI Studio. William Falcon began developing PyTorch Lightning in 2015 at Columbia University. He open-sourced it in 2019 during his PhD at NYU and Facebook AI Research, guided by his advisors Kyunghyun Cho and Yann LeCun. In 2023, we launched Lightning AI Studio, a cloud platform for coding, training, and deploying AI models directly from the browser with zero setup.Today, PyTorch Lightning has over 160M downloads. AI Studio supports 240K+ users across thousands of enterprises.Start a free StudioPyTorch Lightning - 200+ million downloadsLeaders in AI since 2019Our 40+ person team has contributed to and shaped AI projects like PyTorch and PyTorch Lightning, in collaboration with the AI research community. We educate future AI leaders through books, courses, and open-source initiatives.See our platform's evolution:PyTorch Lightning is born2015PyTorch Lightning is bornPyTorch Lightning OSS2019PyTorch Lightning OSS2048 GPU model20192048 GPU modelGPU → TPU (no code changes)2019GPU → TPU (no code changes)Lightning accelerators2020Lightning acceleratorsTorchmetrics2021TorchmetricsLightning Fabric2022Lightning FabricLitGPT2023LitGPTLightning Cloud2023Lightning CloudThunder2024ThunderLitServe2024LitServePartnersOur partnerships with the PyTorch foundation, leading hardware, cloud, and AI labs, empower our users to deploy the fastest, most secure AI models on the world's best infrastructure.NVIDIAMetaAWSIntelGoogleLinkedInPyTorch board memberIndex VenturesCoatueBain Capital VenturesNvidiaCisco InvestmentsJ.P. MorganFirst Minute CapitalSV AngelK5 GlobalBacked by incredible investors$108M from top VCs to empower the next generation of AI builders.Series A announcementSeries B announcement2024 funding announcementJoin our communityOver 3 million developers use Lightning across the world. Join our community. Contribute. Learn from the best. Build AI Lightning fast.Join our DiscordFind an eventNew York·San Francisco·LondonJoin our teamWe're on a mission to empower everyone to build AI.If you move fast, value craftsmanship, think minimally, love focus, and thrive on tough technical challenges, you'll fit right in ⚡️.Join us in New York, San Francisco or London.Explore careers --- ExploreFeaturedTrendingRecentAll templatesEducationalBeginnersBlogsClassesPapersTutorialsTemplatesAgentsAI AppsChatbotsCustom modelsData processingDashboardsEvaluationsInferenceMCP servers NotebooksPipelinesReinforcement learningScienceTrainingWeb appsWorkflowsOtherStart an AI project without setupDescribe what you want, AI builds it for you.Any taskDebugDataTrainOptimizeDeploy (Inference)NotebookAgentDebug my training loop – the loss is NaN after 100 steps.Build itStart from a template​​qrl-qai is the quantum analogue of OpenAI's Reinforcement Learning gym frameworkqrl-qai playground (1.0.0)3 clones46 Views FeaturedZeroClaw + Ollama on Lightning Studio — run a fully autonomous AI agent locally with zero API costs. Pre-installed studio with ~3.4MB Rust runtime, built-in web dashboard, Telegram integration, and local LLM inference. Clone, attach a GPU, and start chatting in minutes.ZeroClaw: The Lightweight OpenClaw Alternative, Powered by Ollama61 clones3.88K Views FeaturedRun Clawdbot instantly on the cloud. No Mac minis, no local installs, no terminal access. Secure, isolated cloud environment with one-click launch and browser-based setup.Clawdbot in the cloud, zero setup (openclaw)997 clones9.05K Views FeaturedCompare open-source OCR models locally using DeepEval and Streamlit. Benchmark Datalab Chandra, DeepSeek OCR, Granite Docling, Qwen-3-VL, and DOTS OCR with LLM-as-a-Judge G-Eval metrics. Upload documents, add ground truth, and visualize performance results.Open-Source OCR Playground46 clones4.12K Views FeaturedBuild scalable agentic reinforcement learning (RL) environments using OpenEnv and Unsloth. Learn to create modular, Docker-based RL setups with standardized Gym-style APIs and memory-efficient LoRA fine-tuning -achieving up to 6× faster training and 70% lower VRAM usage.PyTorch OpenEnv: Environments for Agentic RL training27 clones2.37K Views FeaturedGRPO training implementation for math reasoning using torchforge on 4x H100 GPUs. Trains Qwen3-1.7B on GSM8k with custom rewards for answer correctness and reasoning quality. Features async RL, multi-model orchestration, and real-time metrics tracking.GRPO Training with torchforge on Lightning Studio27 clones805 Views FeaturedRun TorchTitan pre-training using Monarch for interactive, distributed multi-node training on Lightning AI infrastructure.Large-Scale Interactive Training with Monarch13 clones1.41K Views FeaturedReinforcement learning (RL) is becoming a crucial ingredient for aligning and extending large language models (LLMs) beyond what is possible with supervised fine-tuning alone. The veRL framework from ByteDance Seed provides a flexible and high-throughput framework for training LLMs with reinforcemenTraining a Coding Agent with veRL16 clones2.43K Views FeaturedThis studio trains a Deep Q-Network (DQN) for the Atari 2600 games. It uses lightning fabric to accelerate training, uses wandb for logging, and compresses the replay buffer. RL Control on the Arcade Learning Environment (ALE) with DQN6 clones297 Views FeaturedIntegration of OpenSpiel games with the OpenEnv framework, optimized for Lightning Studio development. OpenSpiel is DeepMind's collection of 70+ game environments for RL research.OpenSpiel OpenEnv in Lightning Studio5 clones279 Views FeaturedTrain RL agents for traffic signal control using SUMO OpenEnv in Lightning Studio. Features hot-reload development, configurable rewards, custom network support, and seamless deployment. Optimize traffic light timing to minimize delays in realistic urban scenarios.Traffic control RL environment11 clones344 Views FeaturedQuickstart example to use OpenEnv in Lightning Studios - it shows how to set up isolated execution environments with Gymnasium-style APIs for agentic RL training, including launching environments, testing via API endpoints, and sharing via Docker.OpenEnv RL environments quickstart8 clones916 Views FeaturedNanochat is a minimal, hackable full-stack LLM (ChatGPT-like) in one clean codebase. Runs end-to-end on 8×H100 with scripts like speedrun.sh, covering tokenization, pretraining, finetuning, eval, inference, and web UI chat.Build your own ChatGPT from scratch43 clones3.78K Views FeaturedBuild your own open-source NotebookLM! Upload PDFs, audio, text, or sites and get citation-backed AI answers with memory, podcasts, and full transparency. Powered by OpenAI, AssemblyAI, Firecrawl, Zep, Milvus & Streamlit.Build your own NotebookLM118 clones4.95K Views FeaturedBuild a reasoning LLM with reinforcement finetuning of Qwen 3 using GRPO & Unsloth. Complete local tutorial for math/reasoning tasks with proximity-based rewards & HuggingFace datasets. Includes training & inference notebooks. 100% local setup.Build a reasoning LLM from scratch using GRPO89 clones19.15K Views FeaturedAI Paralegal Assistant for legal research & contract analysis. 32x faster with binary quantization RAG. Analyzes risks, statutes, regulations. Built with CrewAI, Milvus, Firecrawl, OpenAI gpt-oss. Fully local deployment ensures client confidentiality. Transform hours of research into minutes.Multi-agent legal assistant powered by gpt-oss140 clones5.22K Views FeaturedRun OpneAI's GPT-OSS locally with a Thinking UI using Ollama + Streamlit. Enjoy private, fast AI with real-time reasoning, no API fees, and full chat history—experience advanced, open-source LLM performance right on your machine.Run OpenAI gpt-oss locally with Thinking UI57 clones1.10K Views FeaturedInworld Runtime is the first AI runtime for consumer applications. This studio explores the creation of a voice-powered ai shopping agent using the Inworld Runtime. The session will cover key features of the Inworld Runtime, such as its adaptive graphs and automated MLOps.Build Production-ready Voice AI Agents with Inworld Runtime29 clones839 Views FeaturedCompare OpenAI's GPT-5, Anthropic’s Claude Opus 4.1, and other top AI code generation models side-by-side using Opik’s G-Eval metrics. Run models in parallel via LiteLLM & OpenRouter, and get real-time, visual evaluations on correctness, readability, and best practices.Compare OpenAI GPT-5 and Claude Opus 4.1 for code generation10 clones1.14K Views FeaturedBuild a multimodal AI system handling text, images, audio, and video via unified interface. Combines Pixeltable storage with CrewAI orchestration and local LLMs for intelligent content management and semantic search across all media types.Ultimate MCP server for Multimodal AI42 clones2.81K Views FeaturedIn this project, I set out to merge three powerful methodologies—LSTM, XGBoost, and Q-Learning—to decode the chaotic rhythm of the stock market and optimize decision-making under uncertainty.Uncovering Patterns in Stock Market Data: Predictive Analytics and Actionable Recommendations18 clones855 Views FeaturedTransform any text into emotionally rich, human-like speech with Chatterbox TTS, ResembleAI’s state-of-the-art open-source model. With features like emotion exaggeration, zero-shot voice cloning, and neural watermarking, Chatterbox is redefining voice synthesis. Build a Production-Ready TTS API with Chatterbox powered by LitServe153 clones6.67K Views FeaturedWrite books faster with AI, our multi-agent AI book writer powered by CrewAI, local LLMs, and Firecrawl web scraping. The system researches, drafts, and edits chapters autonomously, delivering SEO-ready, plagiarism-free long form content in minutes. That you can publish right away. Build a Multi-Agent Book Writer94 clones4.07K Views FeaturedLearn to build an Agentic RAG app powered by Qwen 3 that searches a vector DB and falls back to web search if needed.Agentic RAG powered by Qwen 393 clones5.43K Views FeaturedAlibaba's Qwen 3 is the latest generation of LLMs in the Qwen series with dense and mixture-of-experts (MoE) models. This studio details the entire process of fine-tuning it 100% locally using Unsloth.Fine-tuning Alibaba's Qwen3 (100% local)149 clones5.64K Views FeaturedLearn to set up a fully local MCP client—no cloud, no data leaks. Step-by-step guide to offline installs, encryption, and airtight security best practices.Build a Private & Secure MCP Client (100% Local)140 clones10.90K Views FeaturedRF-DETR is a groundbreaking, real-time, transformer-based next-generation model developed by Roboflow, pushing the boundaries of real-time performance. This studio provides a comprehensive guide to deploying the state-of-the-art RF-DETR model using LitServe. Deploy RF-DETR: A SOTA Real-Time Object Detection Model using LitServe48 clones10.52K Views FeaturedDiscover a fresh approach to interact with your documents through Google DeepMinds's open source Gemma 3 served locally using Ollama. Query your knowledge base effortlessly to pinpoint exactly what you need using a natural language interface.RAG using Google DeepMind's Gemma 391 clones8.86K Views FeaturedExplore the future of document chat with our Corrective RAG (CRAG) agentic workflow. Powered by DeepSeek-R1, Linkup, and Streamlit, this system autonomously retrieves, evaluates, and enriches document search results—combining local retrieval with web search fallback to ensure high quality response.Build a Corrective RAG Agentic Workflow using DeepSeek-R164 clones6.17K Views FeaturedDiscover the future of automated flight booking with our multi-agent AI system. This solution uses CrewAI, Browserbase, and Ollama to autonomously find and summarize the best roundtrip flights—all while running completely locally to protect your privacy.Build a multi-agent flight booking crew using DeepSeek-R188 clones7.17K Views Featured Discover how to build a cutting-edge Multimodal RAG app using Qwen2.5-VL locally. This guide shows you how to process visually rich documents with ColPali, generate high-quality image embeddings, and store them in a Qdrant vector database.Multimodal RAG using Qwen 2.5 VL-Max312 clones15.78K Views FeaturedBase (minimalist) implementation of https://github.com/comfyanonymous/ComfyUI as a Lightning AI Studio. Just has an upgraded version and the extension manager. This studio is designed for rapid start-up and development of new projects in ComfyUIClean ComfyUI Template v0.3.15 20250221609 clones2.60K Views FeaturedAnalyze and research financials of public listed companies in minutes using an AI research agent. Stock Research Agent with LitServe103 clones18.96K Views FeaturedDeepSeek-R1 delivers OpenAI-o1 level intelligence at 90% less cost. In this studio we build a Streamlit app to compare and evaluate them using RAG. Finally we also do a formal evaluation using an open-source evaluation and observability platform called Opik.Compare DeepSeek-R1 and OpenAI-o1 using RAG92 clones8.72K Views FeaturedDiscover the future of conversational AI with DeepSeek-R1, a state-of-the-art reasoning model. Powered by LitServe and OpenAI-compatible APIs, it offers real-time engagement, quick deployment, and unmatched scalability. Chat with DeepSeek-R1, an Advanced AI Reasoning Model 🤖275 clones19.26K Views FeaturedDiscover a fresh approach to interact with your documents through DeepSeek-R1, served locally using Ollama. Query your knowledge base effortlessly to pinpoint exactly what you need using a natural language interface.RAG using DeepSeek-R1297 clones40.65K Views FeaturedModernGLiNER, a bi-encoder GLiNER model, improves upon traditional unicoder models by enabling unlimited entity recognition, faster inference, and better generalization. This studio demonstrates using the GLiNER model for the Named Entity Recognition (NER) task served using LitServe.Deploy a ModernGLiNER model20 clones2.52K Views FeaturedDiscover how to build and scale a production-ready embeddings API like a pro. Combine LitServe for high-performance infrastructure, OpenAI's Embedding Spec for industry-standard API compatibility, and FastEmbed for efficient embedding generation. This guide provides a step-by-step solution.Build and Scale Embeddings API Like a Pro using OpenAI EmbeddingSpec with LitServe34 clones1.08K Views FeaturedDiscover how to deploy Jina CLIP V2, a state-of-the-art multilingual multimodal embedding model for text and images, using LitServe—a lightning-fast inference engine. Deploy Jina CLIP V2: A Guide to Multilingual Multimodal Embeddings API with LitServe40 clones823 Views FeaturedThis studio shows how to deploy a RAG system on text documents using OpenAI and LitServe.Deploy an OpenAI RAG Server105 clones3.90K Views FeaturedDeploy a voice clone API with F5 text to speech, LitServe and a Lightning StudioDeploy voice clone API (F5 TTS)248 clones6.11K Views FeaturedSpeed up large dataset processing in pandas by 150x by using GPUs with zero code changes. Pandas can be sped up with RAPIDS cuDF. Just load the cudf.pandas extension on a Studio, and it automatically uses GPUs when available, switching to CPUs when necessary. Run pandas on GPUs for a 150x speed up - powered by RAPIDS cuDF36 clones2.83K Views FeaturedThis studio creates a image segmentation API using Litserve and Segment Anything 2, which can extract the subject from a photo. Deploy a image segmentation API with Meta's SAM267 clones2.55K Views FeaturedLearn how to deploy and interact with Llama 3.2-Vision multimodal LLM using LitServe, the fast and flexible FastAPI-based inference engine. Unlock seamless OpenAI compatibility, tool calling, and custom response formats to streamline your AI workflows—all with speed and simplicity.Deploy and Chat with Llama 3.2-Vision Multimodal LLM Using LitServe, Lightning-Fast Inference Engine220 clones5.62K Views FeaturedThis studio shows how to finetune and serve Meta AI's Llama 3.2 language models.Finetune and Serve Llama 3.2 1B and 3B388 clones7.51K Views FeaturedMake your RAG upto 40x faster and 32x memory efficient by leveraging binary quantization. Learn how to search over millions of vectors in milliseconds. Perfect for professionals seeking scalable real-time RAG applications.Deploy a private Llama 3.2 RAG API288 clones12.56K Views FeaturedLearn how to deploy the multi-GPU Jamba 1.5 model as a private API using LitServe, enabling secure and scalable access to its advanced conversational capabilities, optimized for high-performance enterprise applications.Deploy Jamba 1.5 with LitServe31 clones601 Views FeaturedDeploy multimodal Llama 3.2 Vision Instruct model as OpenAI compatible server with LitServeDeploy Llama 3.2 Vision with LitServe98 clones4.29K Views FeaturedDeploy a multi-modal LLM that can process images and text with PixtralDeploy a multi-modal LLM with Pixtral89 clones4.49K Views FeaturedDeploy a fast and efficient Speech Generation API using Parler TTS, powered by LitServe. Generate high-quality text-to-speech audio with low latency, scalable endpoints, and easy integration for your applications."Deploy a Speech Generation API using Parler TTS
 powered by LitServe73 clones19.23K Views FeaturedExplore the deployment and interaction with Qwen2-VL using LitServe’s OpenAI-Compatible API. I’ve walked through deploying the model, querying it with images via a Python client, and even integrated a Streamlit app to chat with Qwen2-VL using images and videos. A seamless approach to harnessing advaDeploy and chat with Qwen2-VL using LitServe171 clones4.96K Views FeaturedDeploy an API that uses both a PyTorch and TensorFlow model in the same endpoint 🤯. LitServe's ability to compose models lets you easily create complex API servers even when the models are written in different frameworks.Deploy both PyTorch and TensorFlow in a single API41 clones2.09K Views FeaturedDeploy an API that generates images from text and rules. This API is powered by ControlNet and LitServe.Deploy a controlled Image Generation API (ControlNet)71 clones2.03K Views FeaturedLearn to deploy Retrieval-Augmented Generation (RAG) applications using LitServe for scalable, serverless access with multi-GPU support. Simplify your AI deployments.Deploy a private Llama 3.1 RAG API168 clones6.86K Views FeaturedThoracic-Surgery-Life-Expectancy-Prediction26 clones577 Views FeaturedDeploy AI Agent with Tool use using OpenAI compatible API format. Deploy AI Agent with Tool Use107 clones3.29K Views FeaturedLearn how to deploy a multi-modal Phi-3.5 Vision Instruct model as a private API with LitServe.Deploy Phi3.5 Vision API with LitServe73 clones2.63K Views FeaturedServe a private text-embedding API for your RAG application. Learn how to deploy a embedding generation API as a private API with LitServe.Deploy text embedding API with LitServe30 clones2.02K Views FeaturedLearn how to deploy a XGBoost model as a private API with LitServe. This tutorial covers real-world use cases like classification, regression, and anomaly detection.Deploy XGBoost with LitServe36 clones2.27K Views FeaturedLearn how to deploy a Scikit-learn Random Forest model as a private API with LitServe. This tutorial covers classification, regression, and anomaly detection using sklearn and LitServe. Deploy random forest with LitServe17 clones2.08K Views FeaturedThe DEEP-EM TOOLBOX supports the application and adaptation of deep learning (DL) solutions in Electron Microscopy (EM) labs, bridging the gap between DL experts and EM researchers. By introducing standardized workflows and plug-and-play use cases, the toolbox fosters interdisciplinary collaborationDEEP-EM TOOLBOX: Segmentation of Cellular Structures18 clones758 Views FeaturedIn this tutorial, you'll learn to train a time series forecasting model using PyTorch Lightning with historical stock price data. We'll leverage a pre-trained sequence model from PyTorch's library, guiding you through dataset setup, model architecture, and training process.Time-Series Forecasting with PyTorch Lightning127 clones7.14K Views FeaturedDeploy a super resolution image API with Aura SR58 clones2.03K Views FeaturedTrain a diffusion model from scratch to generate realistic images. This Studio is used in the README for PyTorch Lightning.Train a diffusion model with PyTorch Lightning 137 clones9.64K Views FeaturedDiscover how to create custom synthetic datasets for preference finetuning with Llama 3.1 & Distilabel to enhance smaller AI models through step-by-step guide in this Studio. Create synthetic datasets with Llama 3.1129 clones4.35K Views FeaturedDeploy an image generation API with Flux, LitServe and Lightning StudiosDeploy an image generation API with Flux554 clones6.85K Views FeaturedDeploy an image generation API with AuraFlow, LitServe and Lightning StudiosDeploy an image generation API with AuraFlow25 clones1.20K Views FeaturedIn this tutorial, you'll learn to train an object detection model using PyTorch Lightning with the WIDER FACE dataset. We'll leverage a pre-trained Faster R-CNN model from torchvision, guiding you through dataset setup, model, and training.Object Detection with PyTorch Lightning75 clones6.36K Views FeaturedThis tutorial provides a comprehensive guide to building a Convolutional Neural Network (CNN) for classifying images of different car brands. It's a minimalistic example using a collected car dataset and standard ResNet architecture.Image Classification with PyTorch Lightning104 clones7.00K Views FeaturedThis studio will guide you through running Meta's Chameleon 30b multi-modal model on cloud GPUs.Run Meta's Chameleon 30b15 clones1.70K Views FeaturedThis studio will guide you through running Meta's Chameleon 7b multi-modal model on cloud GPUs.Run Meta's Chameleon 7b9 clones607 Views FeaturedIn today's data-driven world, processing large datasets securely and efficiently is more crucial than ever. Enter LitData, a powerful Python library that's revolutionizing how we handle and optimize big data. Let's dive into how LitData's advanced encryption features can help you process data at scaUnlock Secure Data Processing at Scale with LitData's Advanced Encryption Features17 clones1.92K Views FeaturedDeploy an noise cancellation API with LitServe, DeepFilterNet and Lightning StudiosDeploy a noise cancellation API with DeepFilterNet17 clones2.56K Views FeaturedThis template uses https://github.com/georgian-io/LLM-Finetuning-Toolkit & vLLM to demonstrate a simple training-to-serving workflow for LLMs.End-to-End LoRA Finetune & Serving Demo69 clones3.29K Views FeaturedThis studio demonstrates Meta's VGG-SfM to build a 3-dimension representation of an object from video or imagesStructure from Motion with Meta's VGG-SfM59 clones2.68K Views FeaturedThis tutorial is aimed at coders interested in understanding the building blocks of large language models (LLMs), how LLMs work, and how to code them from the ground up in PyTorch.LLMs from the Ground Up (Workshop)1.50K clones13.47K Views FeaturedA short example of how you can train a recurrent neural network to generate english text (next-token prediction) using PyTorch Lightning.Train a recurrent neural network with PyTorch Lightning34 clones2.78K Views FeaturedMake your RAG upto 40x faster and 32x memory efficient by leveraging binary quantization. Learn how to search over millions of vectors in milliseconds. Perfect for professionals seeking scalable real-time RAG applications.RAG 40x faster using binary quantization107 clones7.17K Views FeaturedFinetune a simple audio generation model on your own music collection with PyTorch Lightning. This Studio is used in the README for PyTorch Lightning.Finetune a personal AI music generator173 clones5.99K Views FeaturedIn this studio, we will first understand S3D MIL-NCE model by thoroughly reviewing the paper: End-to-End Learning of Visual Representations from Uncurated Instructional Videos. Then, we will walk through a short demo on text-to-video retrieval using pre-trained S3D model.S3D MIL-NCE: Text-to-Video Retrieval32 clones1.39K Views FeaturedDeploy an media conersion API with LitServe, FFmpeg and Lightning StudiosDeploy an media conversion API with FFmpeg38 clones1.91K Views FeaturedGuide for getting started with LitGPT - Load, pretrain, finetune, deploy 20+ LLMs on your own data. Uses state-of-the-art techniques: flash attention, FSDP, 4-bit, LoRA, and more.LitGPT quick start167 clones51.45K Views FeaturedDeploy an music generation API with LitServe, Audio Craft and Lightning StudiosDeploy an music generation API with Meta's Audio Craft65 clones2.18K Views FeaturedThis studio guide walks you through the process of using LitServe on Studios to deploy embedding models, following the OpenAI embedding API format for seamless integration. Notably, this approach supports a wide range of open-source embedding models from Hugging Face, ensuring flexibility.Deploy OpenAI-like Embedding API with LitServe on Studios23 clones1.33K Views FeaturedRun a server that falls back to other model providers when the Open AI API fails.OpenAI fault tolerant proxy server42 clones4.44K Views FeaturedDeploy an audio generation API using LitServe, Stable Audio and Lightning StudiosDeploy an audio generation API with Stable Audio57 clones1.71K Views FeaturedLearn how to deploy Mistral-7B-Instruct-v0.3 and use its function calling features to streamline your tasks. This guide will walk you through the steps to get started, helping you make the most of this advanced AI tool.Function Calling with Mistral-7B-Instruct-v0.3: From Deployment to Execution40 clones2.01K Views FeaturedFinetune a simple text classification model with PyTorch Lightning. This Studio is used in the README for PyTorch Lightning.Text Classification with PyTorch Lightning69 clones4.56K Views FeaturedTrain a simple image segmentation model with PyTorch Lightning. This Studio is used in the README for PyTorch Lightning.Image Segmentation with PyTorch Lightning104 clones6.50K Views FeaturedTrain a simple hello world model with PyTorch Lightning. This Studio is used in the README for PyTorch Lightning.PyTorch Lightning Hello World92 clones4.86K Views FeaturedFinetune a local T5 small (80M) for RAG to do exceptionally well on HotPotQA. Uses only 200 labeled answers. Do this without any hand-written prompts & labels for retrieval or reasoning. Align it easily with your use case.DSPy: Finetune a T5-small to excel at RAG54 clones4.68K Views FeaturedLearn how to track your PyTorch Lightning model training with neptune.ai. Add a few extra lines to your code to visualize and compare your experiments and easily reproduce the results.Track your model training on neptune.ai23 clones1.47K Views FeaturedPhi-3-vision is a lightweight, state-of-the-art 4.2 billion parameter multimodal model with language and vision capabilities, available with a 128k context length.Deploy and chat with Phi-3-vision-128k-instruct189 clones4.22K Views FeaturedUse ML workflows with Kedro in Lighting AI! 🚀 This interactive tutorial environment focuses on reliable MLOps and effective pipeline management. Learn to modularize code, execute step-by-step pipelines, control data dependencies. 🌟Perfect for data enthusiasts aiming to enhance MLOps skills🌟Kedro: An Elegant MLOps Solution41 clones1.32K Views FeaturedDeploy a real-time voice transcription web app using Open AI's Whisper model on cloud GPUs.Deploy a real-time voice transcription app121 clones2.15K Views FeaturedThis studio contains the lab materials from DS-GA 3001.003 Special Topics in DS - Causal Inference in Machine Learning (cross listed also as CSCI-GA 3033.108 Special Topics in CS - Causal Inference in Machine Learning) at New York University in Spring 2024. Causal Inference in Machine Learning - A Course Material at New York University31 clones4.67K Views FeaturedTurn written words into speech that sounds like a particular voice. This Studio deploys a voice mimic API. Uses the Coqui XTTS V2 model.Deploy a voice clone API - Coqui XTTS V2 model591 clones8.05K Views FeaturedDeploy a voice clone API that generates audio spoken by the target voiceDeploy a voice clone API (SVTT2)55 clones2.61K Views FeaturedDeploy a private instance of Llama 3 (8B) on a private, self-managed API with full code access and control.Deploy a private Llama 3 (8B) API282 clones5.84K Views FeaturedLearn how to automate RAG evaluation by generating synthetic test data using ragas, an open-source framework. A step towards building efficient & robust RAG pipelines.Generate synthetic data for RAG evaluation91 clones6.07K Views FeaturedThis project contains a collection of independent, executable scripts that showcase most of the available functionalities in PyTorch Lightning, each covering a new feature or technique. It's organized to help you smoothly progress from basic to advanced PyTorch Lightning concepts.Zero to Lightning: Learning Lightning from Scratch83 clones1.45K Views FeaturedGet up and running quickly with Llama 3 8B with a fully customizable UI in pure PythonReflex Chat App - Llama 3113 clones2.18K Views FeaturedQuery your knowledge base effortlessly without compromising your privacy. Learn how to build a self-hosted document chat RAG application using Cohere's powerful Command R model and a locally served vector database, Qdrant.Self hosted RAG app using Cohere's ⌘R199 clones5.78K Views FeaturedWhere's My Pic is an Image Search Engine powered by CLIP, and it can run locally on your own computer!Where's My Pic? A fully local Image Search Engine!39 clones4.92K Views FeaturedChat with you docs using RAG, featuring newly released Llama-3 & Phi-3. See for yourself which model performs better using an interactive Chat application created using Streamlit.Compare Llama-3 and Phi-3 using RAG262 clones16.95K Views FeaturedDeploy a private instance of Stability AI's Stable diffusion 2 on GPUs.Deploy a private API for Stable diffusion 2223 clones4.05K Views FeaturedThis studio template contains the code to get started with multibackend Keras capabilities. The template is designed to be a starting point for PyTorch users to get started with multibackend Keras capabilities. The template shows how to train a simple Keras model using the PyTorch backend.keras3_with_pytorch_workflow40 clones2.17K Views FeaturedDeploy a private instance of Open AI's Whisper model on GPUs.Deploy a private API for Open AI's Whisper model179 clones4.46K Views FeaturedDeploy ANY Hugging Face model instantly with LitServe. This Studio shows the simple steps to get a Private API endpoint for these models with full code access and control.Deploy a Private API for any Hugging Face model332 clones42.24K Views FeaturedDeploy a private instance of Open AI's CLIP model on a private, self-managed API with full code access and control.Deploy a private API for CLIP model by Open AI39 clones2.60K Views FeaturedDeploy a BERT hugging face model with LitServe. This Studio creates a private, self-managed API with full code access and control.Deploy a Hugging Face BERT model67 clones3.13K Views FeaturedUnleashing the Potential of MONAI Framework for Advanced 3D Lung Tumour Segmentation ResearchEmpowering 3D Lung Tumour Segmentation with MONAI49 clones1.73K Views FeaturedFinetune the optimal model by searching hyperparameters through a Bayesian optimization process.Conditional sweeps8 clones317 Views FeaturedDiscover a fresh approach to interact with your documents through MetaAI's Llama-3, served locally using Ollama. Query your knowledge base effortlessly to pinpoint exactly what you need using a natural language interface.RAG using Llama 3.1 by Meta AI1.15K clones37.68K Views FeaturedGet up and running quickly with Llama 3 8B Instruct in this Ollama powered StudioChat with Meta’s Llama 3 8B189 clones4.87K Views FeaturedLeverage the power of Crew AI to build an intelligent multi agent system. Where each agent has a dedicated task & role. Agents can also interact & delegate task amongst each other. The system we build here scrapes web to automatically write an engaging blog post for you.Build a crew of AI agents371 clones6.58K Views FeaturedDeploy a simple hello world with LitServe. This Studio is used in the README for LitServe.LitServe hello world147 clones2.13K Views FeaturedThis studio is a minimal demo of how to use Google's Gemma 2b (instruct-tuned, v1.1) for summarizing a list of news headlines retrieved from newsapi.org, using HuggingFace Transformers. There is also a minimal web front-end using Gradio. I suggest you try it with 1x T4 GPU for lower latency.Gemma-based summarization of news headlines15 clones753 Views FeaturedDiscover a fresh approach to interact with your documents through Cohere's powerful Command R+ model, specifically crafted for developing RAG applications. Query your knowledge base effortlessly to pinpoint exactly what you need using a natural language interface.RAG using Cohere Command R+403 clones62.80K Views FeaturedBuild a Retrieval-Augmented Generation with Lightning AI, Weaviate, LlamaIndex, Hugging Face, Ollama, and Streamlit.Chat with your code: RAG with Weaviate and LlamaIndex132 clones8.18K Views FeaturedDiabetic retinopathy (DR) remains a significant global health concern, with early detection playing a critical role in preventing vision loss. For those eager to contribute to this vital area of research, a comprehensive project studio is readily available. This studio has already tackled many essential tasks involved in DR detection, providing researchers and enthusiasts with a ready-to-use platform for experimentation.Diabetic Retinopathy Detection: Utilizing Multiprocessing for Processing Large Datasets and Transfer Learning to Fine-Tune Deep Learning Models55 clones2.15K Views FeaturedLeverage LLMs to share engaging commentary on AI content on LinkedIn and Twitter.Social Media Influencer AI-powered bot120 clones2.64K Views FeaturedThis Studio let's you chat with Anthropic AI's Claude 3 family of LLMsTalk with Claude 3217 clones6.33K Views FeaturedExperience a new way to interact with GitHub repositories using our RAG application. Query your codebase using natural language, enhancing productivity and understanding. Chat with your code using RAG!622 clones22.22K Views FeaturedGemma is Google’s latest open-weight LLM. This Studio explains shows you how to use Gemma through Lit-GPT and explains some of the unique design choices of Gemma compared to other LLMs.Understanding, Using, and Finetuning Gemma465 clones24.34K Views FeaturedThis lightning studio contains diffusion model training and inference code that is easy to use, compatible with HF ecosystem and free of boilerplate.Build Diffusion models with PyTorch Lightning & HF diffusers147 clones5.05K Views FeaturedUse AUTOMATIC1111's Stable Diffusion Web UI to generate imagesUse generative AI to create images488 clones3.58K Views FeaturedRun the Google Gemma open models and chat with it in a Streamlit UIRun Google Gemma 2B LLM on Cloud GPUs64 clones2.04K Views FeaturedEfficiently search image datasets with words. This Studio simplifies embedding, indexing, and searching images with user-friendly interfaces, without needing extensive setup. Perfect for anyone looking to leverage their own data.Search any image dataset with words85 clones2.15K Views FeaturedLearn how to take a pretrained model and continue pretraining it on a new dataset of your choice.Continued Pretraining with TinyLlama 1.1B191 clones8.47K Views FeaturedUse, explore, & create from scratch the LAION-400-MILLION images & captions dataset.Download & stream 400M images + text75 clones7.19K Views FeaturedLearn how to format and control LLMs to generate output in a structured format like a JSON, enable the LLMs to use a custom tools. We will enable JSON-only mode and build an AI agent by utilizing external APIs.Structured LLM Output and Function Calling with Guidance41 clones6.25K Views FeaturedUse this Studio as a simple tutorial to convert geospatial data into our streaming formatConvert spatial data to Lightning Streaming29 clones1.66K Views FeaturedModel merging is a technique for combining multiple pretrained or finetuned LLMs into a single, more powerful model. This approach is particularly useful when individual models excel in different domains or tasks, and merging them can create a model with a broader range of capabilities and improved overall performance.Efficient Linear Model Merging for LLMs58 clones8.87K Views FeaturedChat with Allen Institute for AI's OLMo 7BRun Allen Institute for AI OLMo 7B9 clones1.92K Views FeaturedSheepRL Workshop: add Super Mario Bros environment and train an agent on it.SheepRL: How to integrate Super Mario Bros enviroment31 clones1.29K Views FeaturedRecipe for instruction finetuning TinyLlama 1.1B on a single GPU in under 10 minutes and for less than 5 credits.Instruction finetuning - TinyLlama 1.1B LLM571 clones6.76K Views FeaturedRun CodeLlama 70B Instruct and chat with it in a Streamlit UIRun CodeLlama 70B Instruct125 clones2.37K Views FeaturedChat with your documents with RAG. AI-Powered PDF Assistant! Upload PDFs, Ask Questions, and Unlock Instant Answers with Language Model. Powered by Langchain Expression Language (LCEL)Document Chat Assistant using RAG382 clones11.83K Views FeaturedProxy-tuning offers a way to adapt LLMs without changing the model's weights. This is especially attractive if a given LLM is too resource-intensive to train, or if a user doesn't have access to the LLM's weights.Improve LLMs With Proxy-Tuning56 clones9.64K Views FeaturedUse this Studio to generate embeddings for the entire 6M English articles from Wikipedia. You can easily adapt the code to embed your own articles.Embed English Wikipedia under 5 dollars46 clones4.18K Views Featuredfinetune Google's BERT model for sequence classification with PyTorch Lightning.Finetune Hugging Face BERT with PyTorch Lightning210 clones6.18K Views FeaturedIngest documents (text, pdf, markdown, docx) in a vector database for Retrieval Augmented Generation (RAG)Document Search and Retrieval using RAG1.60K clones14.21K Views FeaturedUse this Studio to benchmark several cloud data-loading frameworks such as PyTorch Lightning Data, Webdataset, and Mosaic ML. Benchmark cloud data-loading libraries59 clones5.44K Views FeaturedThis Studio explains how to use or prepare TinyLLaMA dataset to pretrain LLMSPrepare the TinyLlama 1T token dataset122 clones4.78K Views FeaturedLoRA (Low-Rank Adaptation) is a popular technique to finetune LLMs more efficiently. This Studio explains how LoRA works by coding it from scratch, which is an excellent exercise for looking under the hood of an algorithm.Code LoRA from Scratch538 clones47.27K Views FeaturedBenchmark vLLM - 2x faster LLM Inference with AWQ (4-bit quantization) for Mistral 7B using vLLM. Learn how to serve open LLMs with OpenAI API protocol and run quantized LLMs in production.Optimized LLM inference API for Mistral 7B using vLLM124 clones17.06K Views FeaturedTrain the TinyLlama language model from scratch. This studio has everything you need, including the training data of about 1.2 TB.Pretrain LLMs - TinyLlama 1.1B315 clones11.87K Views FeaturedRun the Mixtral LLM and chat with itRun Mistral MoE (Mixture of experts)398 clones3.45K Views FeaturedExperiment with Stable Diffusion Pipelines using a node-based visual editor right in your browser with this LightningAI implementation of https://github.com/comfyanonymous/ComfyUI (commit #c782144).Stable Diffusion with ComfyUI4.83K clones275.92K Views FeaturedServe Image Classification model using FastAPI for free. Learn to host Python web apps with FastAPI and Lightning Studio ⚡️Deploy Machine Learning API with FastAPI for free112 clones6.10K Views FeaturedIn this quest, we'll finetune Open AI's CLIP model to a small dataset. We'll start on a CPU Studio and use a GPU to 36x the speed. We'll also monitor the finetuning with Tensorboard and share the public link with colleagues.Finetune a pretrained model4.33K clones15.52K Views FeaturedFinetune the optimal model by trying different hyperparameter combinations using grid search or random search.Run a hyperparameter sweep2.67K clones8.41K Views FeaturedWe'll use CLIP, a model that understands images based on language, and host a streamlit app.Host an AI web app with OpenAI's CLIP model8.85K clones24.03K Views FeaturedClip is a model that's really good at understanding and recognizing images based on languageOpen AI Clip - Describe images with text260 clones5.03K Views FeaturedFinetuning on public and custom datasets for the CodeLlama 13B large language modelFinetune LLM - CodeLlama 13B355 clones1.96K Views FeaturedInference server for the CodeLlama 13B large language modelDeploy LLM Chat API - CodeLlama 13B235 clones2.06K Views FeaturedFinetuning on public and custom datasets for the Llama2 7B large language modelFinetune LLM - Llama2 7B505 clones2.63K Views FeaturedInference server for the CodeLlama 7B large language modelDeploy LLM Chat API - CodeLlama 7B135 clones1.80K Views FeaturedFinetuning on public and custom datasets for the Llama2 13B large language modelFinetune LLM - Llama2 13B116 clones1.22K Views FeaturedInference server for the Llama2 13B large language modelDeploy LLM Chat API - Llama2 13B54 clones1.30K Views FeaturedFinetuning on public and custom datasets for the CodeLlama 7B large language modelFinetune LLM - CodeLlama 7B337 clones3.23K Views FeaturedInference server for the Llama 7B large language modelDeploy LLM Chat API - Llama2 7B150 clones2.43K Views FeaturedInference server for the Mistral 7B large language modelDeploy LLM Chat API - Mistral 7B1.05K clones5.97K Views FeaturedFinetuning on public and custom datasets for the Mistral 7B large language modelFinetune LLM - Mistral 7B558 clones4.81K Views Featured