accelerate

6719cb6d - Avoid duplicating memory for tied weights in `dispatch_model`, and in forward with offloading (#2330)

Avoid duplicating memory for tied weights in `dispatch_model`, and in forward with offloading (#2330) * wip * fix * add test * cleanup * style * style & tests pass * fix offload, submodules * cleanup * Update tests/test_big_modeling.py Co-authored-by: Marc Sun <57196510+SunMarc@users.noreply.github.com> * Update tests/test_big_modeling.py Co-authored-by: Marc Sun <57196510+SunMarc@users.noreply.github.com> * disk offloading do not reload tied parameters in memory * remove outdated comment --------- Co-authored-by: Your Name <you@example.com> Co-authored-by: Marc Sun <57196510+SunMarc@users.noreply.github.com>