Make your data ready.
Query and edit text datasets. Create object classes, draw masks and bounding boxes, and export labeled images.
Coming soon for Mac
Prepare datasets, fine-tune LLMs and diffusion models, and create adapters in one visual workspace.
From raw data to a checkpoint you can test.
The whole workflow
Move from the first row or image to a trained model, without losing the thread of your experiment.
Query and edit text datasets. Create object classes, draw masks and bounding boxes, and export labeled images.
Fine-tune LLMs and diffusion models or create adapters. Choose a supported local model, configure a run, and follow its progress.
Try a saved checkpoint in the app. Chat with your LLM adapter or generate an image from your diffusion checkpoint.
01 / Fine-tuning
Fine-tune LLMs and diffusion models with your own datasets. Choose a supported local model, set your training parameters, and follow the run.
Train a language model for your task with prepared text and conversation data. Configure the run and review its progress and checkpoints.
Adapt an image model to your dataset. Select the model and its components, follow training, then generate an image to verify a saved checkpoint.
Choose a local diffusion model and its components, load the NYC dataset, configure training, and generate an image from the saved checkpoint.
Unable to play the video. Open it directly.
Actual app recordings. Walkthroughs are edited for length; training waits are shortened.
02 / Adapter creation
Create adapters for LLMs and diffusion models. Bring a prepared dataset and a supported base model, then configure the adapter for your experiment.
Load a prepared LLM dataset, configure adapter settings, train the adapter, and verify the saved checkpoint in chat.
Unable to play the video. Open it directly.
Adjust adapter settings, start training, and test a saved checkpoint by chatting with it in the app.
Teach an image model a subject or style. Train a compatible adapter and explore its effect through image generation.
03 / LLM dataset preparation
Bring text and conversations into one workspace. Find the rows you need, refine the details, and build the dataset for your next run.
Load local data, run a SQL query, edit a record, and generate five new rows.
Unable to play the video. Open it directly.
04 / Image dataset preparation
Three ways to prepare image data: label regions, draw bounding boxes, or curate images and captions for generative models.
Create object classes, draw masks around individual objects or pixel regions, review each image, and export your segmentation dataset.
Load the NYC images, create car and people classes, mask objects in both images, and export all prepared data.
Unable to play the video. Open it directly.
Define classes and draw bounding boxes. Label cars, people, or the objects your model needs to recognize, then export the prepared data.
Load the NYC images, create car and people classes, draw bounding boxes in both images, and export all prepared data.
Unable to play the video. Open it directly.
Curate images and captions for diffusion fine-tuning or adapter creation. Set training splits, review optional training regions, and generate new images with a supported local model.
A few things to know
A focused workspace for hands-on AI development on your Mac.
ML developers, researchers, and dataset creators who want a visual workspace for preparing data, annotating images, fine-tuning models, creating adapters, and checking results.
Prepare LLM datasets and image datasets for segmentation, object detection, and image generation. Fine-tuning and adapter workflows cover supported LLMs and diffusion models. Available training methods depend on the selected model and its format.
Yes. You can select supported local models and configure the components a workflow needs. The demos show an LLM adapter with SmolLM2 and diffusion fine-tuning with Stable Diffusion 1.5. Compatibility depends on the model, runtime, and your Mac.
A compatible Mac, a supported model, a prepared dataset, and a configured Python environment. Flask Workspace includes setup checks for training. Model size and training settings affect memory, storage, and run time.
Flask Workspace is coming soon for Mac. The download link will appear on this page when the app is available. Until then, explore the recorded workflows above.