Skip to content

Force MHA QKV onto fp32 #5395

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
titu1994 merged 1 commit into from
Nov 10, 2022
Merged

Force MHA QKV onto fp32 #5395

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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: 5 additions & 0 deletions nemo/collections/asr/parts/submodules/multi_head_attention.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -142,6 +142,8 @@ def forward(self, query, key, value, mask, pos_emb=None, cache=None, cache_next=

if torch.is_autocast_enabled():
query, key, value = query.to(torch.float32), key.to(torch.float32), value.to(torch.float32)

# temporary until we solve this more gracefully
with avoid_float16_autocast_context():
q, k, v = self.forward_qkv(query, key, value)
scores = torch.matmul(q, k.transpose(-2, -1)) / self.s_d_k
Expand Down Expand Up @@ -218,6 +220,9 @@ def forward(self, query, key, value, mask, pos_emb, cache=None, cache_next=None)
"""
key, value, query = self.update_cache(key=key, value=value, query=query, cache=cache, cache_next=cache_next)

if torch.is_autocast_enabled():
query, key, value = query.to(torch.float32), key.to(torch.float32), value.to(torch.float32)

# temporary until we solve this more gracefully
with avoid_float16_autocast_context():
q, k, v = self.forward_qkv(query, key, value)
Expand Down