Skip to recipe workspace
B70 Launcher
Connecting
Usage
Choose a model
Connecting to your local launcher.
Configuration
Context tokens
GPUs
Server port
Target application
Advanced
KV cache precision
Prefill
Prefix cache
Off is for testing and benchmarks.
Parallel slots
Speculative draft (MTP/DFlash), when available
Extra engine flags
Environment variables
KEY=VAL, one per line.
Recipe details
Confirm download
Download files
Cancel
Change folder
Open the model page on Hugging Face
Connecting
Preview command
Close