lebowski model (#38268)

* Reapply "usbgpu: never materialize model weights in RAM" (#38271)

This reverts commit 8196b743af.

* release oob buffers as we load them

* don't leak fds

* lint

* bigger big

* don't roundtrip the features to qcom

* Revert "don't roundtrip the features to qcom"

This reverts commit 0d2ca73d7a71bd32c46beda425957339f1dbbcf8.

* tmp bump allowed lag

* faster

* Revert "tmp bump allowed lag"

This reverts commit b407702a0242074eb12abdff14e61789fc05109a.
This commit is contained in:
Armand du Parc Locmaria
2026-07-01 01:26:11 -07:00
committed by GitHub
parent a2c554805b
commit fa0c6876d3
@@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:471f372751d5931e939320c211e35b7255f8fa8015125b3b3fd48ef43020257e
size 195490097
oid sha256:a501760a9d1d5fef0eab2b8c5d122d06124fc26dc8e0782e0aa94b82a208f0ff
size 1757355221