Skip to content

Fix GPT Neo-X multi-gpu inference #2401

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

Merged
jeffra merged 6 commits into from
Oct 13, 2022
Merged

Fix GPT Neo-X multi-gpu inference #2401

Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
248d63e
Fix GPT Neo-X multi-gpu inference
andrewchernyh Oct 7, 2022
6a02075
Merge branch 'master' into fix-gpt-neox-mgpu-inference
RezaYazdaniAminabadi Oct 7, 2022
087418c
Merge branch 'master' into fix-gpt-neox-mgpu-inference
RezaYazdaniAminabadi Oct 11, 2022
a67efc1
Merge branch 'master' into fix-gpt-neox-mgpu-inference
RezaYazdaniAminabadi Oct 13, 2022
8e58fc8
Merge branch 'master' into fix-gpt-neox-mgpu-inference
jeffra Oct 13, 2022
daacdfc
Merge branch 'master' into fix-gpt-neox-mgpu-inference
jeffra Oct 13, 2022
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
5 changes: 2 additions & 3 deletions deepspeed/module_inject/replace_module.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -454,9 +454,8 @@ def transpose(data):
dense_w.data = transpose(dense_w.data)

def _transpose(x):
num_attention_heads_per_partition = transformer_config.heads // transformer_config.mp_size
attention_head_size = x.shape[-1] // num_attention_heads_per_partition
new_x_shape = x.size()[:-1] + (num_attention_heads_per_partition,
attention_head_size = x.shape[-1] // transformer_config.heads
new_x_shape = x.size()[:-1] + (transformer_config.heads,
attention_head_size)
x_1 = x.view(*new_x_shape)
(q, k, v) = torch.split(x_1, (x_1.shape[-1] // 3), dim=(x_1.dim() - 1))
Expand Down