Local LLM / Local memory and quantization
Local LLM: Out-of-memory recovery
Reject or retry with an explicit reduced resource profile instead of repeating the identical request indefinitely.
Environment and inputs: Local inference capacity and quantization; supported precision and offload behavior are runtime- and hardware-specific.
When to use it
Model file size is only part of the memory budget. Measure the serving process with realistic context lengths and active sequences, and keep a quality baseline so a memory saving does not silently change required task behavior.
Check the result
Force an allocation failure and verify bounded recovery and clear user status.
Save and compare results
Save the test input, configuration and result. Fix the failing boundary, repeat the same test and compare it with the saved working configuration.
Scope and limits
These exercises have not been tested on your systems. Check feature support and configuration names against the installed release and linked documentation.