Tokenshark
+
Weights
auto
all on GPU/CPU
RAM → device per layer
stream from disk
keep
Load
Unload
🗑
…
🐞 Step
Block ▸
Token ▸▸
Finish
✕
▶ Run
Tasks
⏸
■
The app was updated on disk but this server is still running the old code. Close the run.bat window (Ctrl+C) and start it again, then reload this page.
Add a model
Id (short name)
Family
Text → text (Llama / Qwen / Mistral / mixture-of-experts)
Text → image (Stable Diffusion 1.x/2.x)
Hugging Face repo (e.g. Qwen/Qwen2.5-3B-Instruct)
…or a local folder (config.json + *.safetensors, or Meta params.json + consolidated.00.pth)
Size (GB, for display)
Notes
Add
Cancel