Skip to content

Assert mp_size is factor of model dimensions #2891

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 7 commits into from
Mar 21, 2023
Merged

Assert mp_size is factor of model dimensions #2891

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
0f11aa1
Enforce mp_size is factor of model parameters such as hidden size
molly-smith Feb 24, 2023
5246b94
Merge branch 'master' into mosm/mpsize_check
molly-smith Feb 28, 2023
ce0c413
Merge branch 'master' into mosm/mpsize_check
molly-smith Mar 2, 2023
e92e13f
include param sizes in assert message
molly-smith Mar 10, 2023
150ddb6
Merge branch 'master' into mosm/mpsize_check
molly-smith Mar 10, 2023
a5976c1
Merge branch 'master' into mosm/mpsize_check
jeffra Mar 21, 2023
96e5a7b
change test name for amd
jeffra Mar 21, 2023
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
2 changes: 1 addition & 1 deletion .github/workflows/amd.yml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -15,7 +15,7 @@ concurrency:
cancel-in-progress: true

jobs:
unit-tests:
amd-tests:
# The type of runner that the job will run on
runs-on: [self-hosted, amd]

Expand Down
8 changes: 8 additions & 0 deletions deepspeed/module_inject/replace_module.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -464,20 +464,28 @@ def _slice_embedding(child, name, conv_linear_layer):

def update_mp_params(child):
if hasattr(child, 'n_heads'):
assert child.n_heads%mp_size == 0, "n_heads ({}) must be divisible by mp_size ({})".format(child.n_heads, mp_size)
child.n_heads = child.n_heads // mp_size
if hasattr(child, 'inner_dim'):
assert child.inner_dim%mp_size == 0, "inner_dim ({}) must be divisible by mp_size ({})".format(child.inner_dim, mp_size)
child.inner_dim = child.inner_dim // mp_size
if hasattr(child, 'num_heads'):
assert child.num_heads%mp_size == 0, "num_heads ({}) must be divisible by mp_size ({})".format(child.num_heads, mp_size)
child.num_heads = child.num_heads // mp_size
if hasattr(child, 'num_attention_heads'):
assert child.num_attention_heads%mp_size == 0, "num_attention_heads ({}) must be divisible by mp_size ({})".format(child.num_attention_heads, mp_size)
child.num_attention_heads = child.num_attention_heads // mp_size
if hasattr(child, 'num_attn_heads'):
assert child.num_attn_heads%mp_size == 0, "num_attn_heads ({}) must be divisible by mp_size ({})".format(child.num_attn_heads, mp_size)
child.num_attn_heads = child.num_attn_heads // mp_size
if hasattr(child, 'all_head_size'):
assert child.all_head_size%mp_size == 0, "all_head_size ({}) must be divisible by mp_size ({})".format(child.all_head_size, mp_size)
child.all_head_size = child.all_head_size // mp_size
if hasattr(child, 'embed_dim'):
assert child.embed_dim%mp_size == 0, "embed_dim must ({}) be divisible by mp_size ({})".format(child.embed_dim, mp_size)
child.embed_dim = child.embed_dim // mp_size
if hasattr(child, 'hidden_size'):
assert child.hidden_size%mp_size == 0, "hidden_size ({}) must be divisible by mp_size ({})".format(child.hidden_size, mp_size)
child.hidden_size = child.hidden_size // mp_size

conv_linear_layer = False
Expand Down