Increase autofitted context length by dynamically adjusting prefill step size (#367)
* Add dynamic per-request MLX prefill plans * Use full memory-safe dynamic context * Document prefill chunk limits * Update merged vision kit test setup * Test dynamic Gemma image prefill integration * Document Gemma image plan crossings
N
Neil Mehta committed
bc4bd41f908c2dac15ea78d2f714bdcff3b64009
Parent: 3b6d614
Committed by GitHub <noreply@github.com>
on 8/21/2026, 3:00:02 PM