The selected runtime family documents this operating-system path, but version and backend still require a local test.
Inference ≠ training
Estimate the fit.
Then test the device.
Memory math can reject obvious mismatches and identify borderline configurations. It cannot certify drivers, model architectures, prompt templates, runtimes, context implementations, or training toolkits.
Check inference and training separately.
This calculator estimates model-weight, context, and runtime memory. It cannot certify a model, driver, prompt template, or toolkit.
Estimated working memory: 6.4 GB including approximately 4.4 GB of weights and 0.8 GB of context allowance.
Turn on the training estimate only if a compatible toolkit and trainable base model are available.
Allows model files, temporary data, logs, and one additional model copy. Toolkit behavior can require more.