Skip to content

Fix issue with RNG state shape #2164

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
wants to merge 4 commits into from
Closed

Fix issue with RNG state shape #2164

Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
9a9efa9
Fix issue with RNG state shape
epwalsh Sep 8, 2025
2647405
clean up
epwalsh Sep 8, 2025
97687ed
Merge branch 'main' into epwalsh/2159-patch
phu0ngng Sep 17, 2025
833dcc0
Merge branch 'main' into epwalsh/2159-patch
phu0ngng Sep 29, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
3 changes: 1 addition & 2 deletions transformer_engine/jax/cpp_extensions/attention.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1820,8 +1820,7 @@ def ring_attn_fwd_impl(

# RNG shape should be the shared shape. This is unused for ring attention as we do not
# support dropout currently.
rng_state_shape = (result_infos[2].shape[0] // mesh.size, *result_infos[2].shape[1:])
rng_state = jnp.zeros(rng_state_shape).astype(result_infos[2].dtype)
rng_state = jnp.zeros(result_infos[2].shape).astype(result_infos[2].dtype)

def scan_kv_block(idx, carry):
kv, output, softmax_aux = carry
Expand Down
Loading