Skip to content

Set accumulate type to bf16 in activation quant #152

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
merged 2 commits into from
Jul 19, 2024
Merged

Set accumulate type to bf16 in activation quant #152

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
036e251
set accumulate type to bf16
lsy323 Jul 19, 2024
df9fd38
fix comment
lsy323 Jul 19, 2024
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: 3 additions & 2 deletions jetstream_pt/layers.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -139,7 +139,8 @@ def forward(self, inputs):
if not self.quantize_activation:
result = F.linear(inputs, self.weight)
else:
# We have to call jax because we need to do dot(int8, int8)->int32.
# We have to call jax because we need to specify the output dtype of dot
# dot(int8, int8)->bf16.
# This semantic cannot be represented in torch. The inferred output dtype
# will be int8 in torch, causing the dot result to overflow.
result = torchjax.call_jax(
Expand All @@ -148,7 +149,7 @@ def forward(self, inputs):
self.weight,
(((2,), (1)), ((), ())),
None,
jnp.int32.dtype,
jnp.bfloat16.dtype,
)
result = result * self.weight_scaler
if self.quantize_activation:
Expand Down
Loading