Transformer Lab

We’ll be attending and exhibiting at Mozilla Festival 2025 in Barcelona (Nov 7–9). It’s a gathering of people building a more open, inclusive internet. Ali will be speaking on (Re)Wiring Growth While Maintaining Open Source Values (session details: https://lnkd.in/ge7ss-DD). Come by our booth and say hello (Room E1 - Mozilla Festival Social Hub). We’ll be answering your questions on how to better train models and run your GPU workloads. Reach out if you're attending! #MozFest

We've talked to many researchers at universities, AI labs and enterprises over the past year. They describe a consistent challenge: as models, datasets, and compute footprints grow, the practical work of running experiments becomes harder and more fragmented. Managing hardware, environments, and orchestration layers often takes as much care as designing the models themselves. The infrastructure layer of machine learning has become the hidden bottleneck for progress. Frontier labs solved this by investing enormous effort into building and maintaining internal tooling to get experiments running reliably at scale. It’s powerful work, but it fragments the ecosystem and keeps the best tools locked inside individual labs. That’s why we’re building a new kind of open-source system, a Machine Learning Research Platform. It helps engineers run distributed training across clusters, manage experiments, and version datasets without needing to maintain complex infra. Our users have started calling it “the essential open-source stack for serious ML teams.” Here’s our vision for how this platform can accelerate research for everyone: 👉 https://lnkd.in/ggDzEJMA Need help building ML tooling to overcome a specific challenge? Please reach out. We’re looking for design partners.

🚀 Launch Announcement: Transformer Lab GPU Orchestration Whether you’re a research lab with racks of GPUs, a startup shipping AI products or a solo AI/ML engineer, you’ve faced similar challenges: outdated SLURM capabilities, compute bottlenecks, and fragmented non-standardized tooling. We’ve been building an open source machine learning research platform (MLRP) to automate away the infrastructure headaches so you can iterate faster and focus on your research. Today is a big leap forward! We're announcing the launch of Transformer Lab GPU Orchestration, a flexible open source platform for AI/ML teams to manage large-scale training across clusters of GPUs. It handles your distributed training orchestration with unified resource pools across on-premise and cloud, automatic job routing, and GPU scheduling designed for modern AI/ML workloads. A modern SLURM replacement built on SkyPilot, Ray, and Kubernetes. Researchers can now move seamlessly from a laptop experiment to a thousand-GPU run without rewriting orchestration scripts or hiring a dedicated infrastructure team. Plus, it’s got built-in support for training and fine-tuning language models, diffusion models, and text-to-speech with integrated checkpointing and experiment tracking. If you’re a university, research institute or AI/ML company interested in a better alternative to SLURM, sign up for our beta. It’s easy to pilot alongside your existing SLURM implementation. Links in comments.

🚀 New Video Drop! I just published a walkthrough on Transformer Lab — an open-source platform for training, fine-tuning, and evaluating advanced LLMs, diffusion, and audio models. 💻 It runs locally on Windows, macOS, and Linux with NVIDIA, AMD, and Apple silicon support — plus you can spin it up easily on RunPods GPU. In this video, you’ll learn: ✅ What Transformer Lab is and why it’s powerful ✅ How to install it locally & on RunPods ✅ How to download and run models ✅ Training & fine-tuning LLMs ✅ Using plugins to extend functionality ✅ Creating datasets for custom training Whether you’re a researcher, developer, or just exploring AI, this tool gives you everything you need to train, evaluate, and deploy AI models with ease. 👉 Watch here: https://lnkd.in/dCm9ZB-7 👉 Try RunPods GPU: https://get.runpod.io/pe48 👉 Transformer Lab docs: https://lnkd.in/dqXBStcZ Let’s push the boundaries of open-source AI together! #AI #LLM #TransformerLab #OpenSource #MachineLearning #DeepLearning #GenAI #AITools #FineTuning #ModelDeployment

We just launched audio modality support for Transformer Lab. It’s the easiest way to generate, clone, and train voices. - Turn text into speech (TTS) - Fine-tune open source TTS models on your own dataset - Clone a voice in one-shot from just a single reference sample - Open source and runs locally on NVIDIA, AMD and Apple Silicon The only platform where you can train text, image and speech generation models in a single modern interface. Speech-based applications are taking off. More and more, teams are preferring privacy, cost effectiveness and don’t want to be locked into a proprietary provider. We’d love your feedback. If you’re experimenting with open source speech models, come build with us. Here's a link to get started along with helpful examples: https://lnkd.in/gBFyMY6r

Every model encodes biases and preferences. Log probabilities (logprobs) help peek under the hood and see what your model really thinks. The NYU MLL CrowS-Pairs dataset has a helpful set of questions to evaluate model bias. Fascinating to run these prompts across models of different origins (Qwen, LLaMA, and Mistral). It’s a reminder that models reflect the training choices behind them. What biases have you observed? Test drive the logprob visualizer in Transformer Lab (transformerlab.ai) yourself! Easy to use and open source.

Are you tired of fine-tuning by writing boilerplate code? You are not alone. When I was looking for a tool/platform to make my life easier with fine-tuning LLMs/LVMs, I came across Transformer Lab It is a leading open source platform to fine-tune advanced AI models without writing a single line of code! Take a look at our video walk through of the Transformer Lab platform demonstrating the fine-tuning process: Video: https://lnkd.in/eY6g6myf Transformer Lab: https://transformerlab.ai Get started here: https://lnkd.in/e53bVut5

🚀 We just launched hyperparameter sweeps. It’s the easiest way to implement sweeps to find the best hyperparameters for LLM training. No more guessing the best config. Instead of adjusting learning rates, batch sizes, and other parameters manually, you give it a set of values and let sweeps explore them automatically. The visualizations make it easy to see which configurations actually improved performance. Works with: ✅ Local setups (CUDA, ROCm, MLX) ✅ Popular model architectures for fine-tuning ✅ Models of any size your hardware can handle Save the headaches and get to the best configurations faster. Best of all, we’re open source (AGPL-3.0). Start using Transformer Lab if you’re interested in leveling up your AI/ML engineering best practices. 🔗 Try Transformer Lab → https://transformerlab.ai/  ⭐ Useful? Give us a star on GitHub → https://lnkd.in/gCWVaNfu #MachineLearning #MLOps #ModelTraining #OpenSource

🎉 Exciting Announcement: Transformer Lab Now Supports AMD GPUs! 🎉 We're thrilled to announce that Transformer Lab now fully supports AMD GPUs on both Linux and Windows platforms! This expansion marks a significant milestone in our mission to make advanced AI tools more accessible to the broader developer community. Whether you're running Linux natively or using Windows with WSL, you can now leverage your AMD hardware to run and train transformer models with Transformer Lab. Behind the Scenes Adding AMD support was no small feat. Our team: * Built a dedicated AMD test rig from scratch * Navigated the complexities of ROCm (AMD's equivalent to CUDA) * Solved critical implementation challenges with PyTorch on WSL * Developed automated solutions for common setup issues One of our proudest achievements was developing an automatic fix for PyTorch detection issues on WSL, eliminating a major pain point for Windows/WSL users. No manual workarounds needed—everything just works out of the box! What This Means For You If you've been waiting to use Transformer Lab with your AMD hardware, that wait is over. This update opens new possibilities for: * Researchers working with limited budgets * Developers with AMD-based workstations * Teams looking to diversify their hardware stack Our comprehensive installation guide makes getting started a breeze: https://lnkd.in/gXu7F7rH Looking Ahead While we encountered some limitations (such as bitsandbytes compatibility issues with ROCm), we're committed to continuing improvement. This is a key step in our journey to make advanced machine learning available to *everyone*. We'd like to extend special thanks to the ROCm community and contributors who helped make this possible. Are you an AMD user excited to try Transformer Lab? Or do you have questions about compatibility with your specific setup? Let us know in the comments below! #TransformerLab AMD #AI #DeepLearning PyTorch

Announcing Qwen3 Support in Transformer Lab. Restart Transformer Lab to get updates for Qwen3 Support in: - MLX Inference and Training - CUDA Inference and Training (including SFT, LoRA, DPO/ORPO/SIMPO) - MultiGPU CUDA Training - GGUF + MLX Export - Reward Modelling... - ++