MyML — Run a model with the hood open
MyML runs real GGUF language models (Qwen3 0.6B, Llama 3.2 1B Instruct, Qwen3 1.7B, SmolLM3 3B, or any compatible Hugging Face GGUF) directly on your GPU in the browser using WebGPU—100% locally with zero servers.
What you can do in MyML
- Steer live behavior dials, conditional CAST triggers, and 3-column Dial Spectrum Sweeps.
- Click any generated word to fork the reply from its top 5 runner-up probabilities.
- Inspect every layer with Logit Lens, Direct Logit Attribution, layer skip/loop, and head muting.
- Run mechanistic interpretability tools: Concept Neurons, SVD Weight Axes, Word Math, Activation Transplant, Wanda Pruning, 2D XY Blending, SAE Feature Clamping, Entropix Sampling, Word Ban/Magnet, and in-browser Rank-4 LoRA training.
- Learn every lever in plain English via 69 ELI5 topics and inline tooltips.
Enable JavaScript in a WebGPU-capable browser to launch MyML, or read /llms.txt for the full machine-readable overview.