George HotzandGitHub 5524916e39 llama compute gradients explicitly + 243 GB of RAM on MP=8 (#15343) ...

* llama compute gradients explicitly

* apply grads

* fix multi issue

* multi BUFFER_VIEW support

* simpler

* skip the flaky test

2026-03-18 19:54:40 +08:00

..

__init__.py

update docs, remove corealize (#4264)

2024-04-23 12:05:29 +04:00

allocations.py

llama compute gradients explicitly + 243 GB of RAM on MP=8 (#15343)

2026-03-18 19:54:40 +08:00

jit.py

Revert "remove mop_cleanup [pr] (#15217)" (#15218)

2026-03-11 04:17:56 -04:00

memory.py

memplanner opt copy bufs (#15110)

2026-03-08 22:28:01 +03:00

realize.py

rm ops.encdec (#15193)

2026-03-09 18:52:48 +03:00

schedule.py

UOp.assign is store+after [pr] (#15292)

2026-03-16 04:51:50 -04:00