Skip to content
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

Fix DML cache generation for 128k models #661

Merged
merged 2 commits into from
Jul 4, 2024
Merged

Fix DML cache generation for 128k models #661

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
7ca7818
Fix DML cache generation for 128k models
PatriceVignola Jun 28, 2024
58a5463
Merge branch 'main' into user/pavignol/fix-dml-cache-generation
natke Jul 3, 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
10 changes: 9 additions & 1 deletion src/python/py/models/builder.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -897,11 +897,20 @@ def make_rotary_embedding(self, rotemb, name, root_input, **kwargs):
def make_rotary_embedding_multi_cache(self):
# Create dummy rotary embedding class
rotemb = type("RotaryEmbedding", (object,), {'content':{}})()
if_cos_cache_output, if_sin_cache_output = "cos_cache", "sin_cache"

# Create caches for when sequence_length > self.original_context_length
self.rotemb_attrs["rescale_factors"] = self.rotemb_attrs["multi_cache"]["long_factor"]
self.rotemb_attrs["cache_length"] = self.context_length
self.rotemb_attrs["mscale"] = self.rotemb_attrs["multi_cache"]["long_mscale"]

# DML doesn't support dynamic selection of the cos/sin cache, so we always use the biggest one
if self.ep == "dml":
self.make_rotary_embedding_caches(rotemb)
self.make_value_info(if_cos_cache_output, self.io_dtype, shape=["max_sequence_length", "head_dim / 2"])
self.make_value_info(if_sin_cache_output, self.io_dtype, shape=["max_sequence_length", "head_dim / 2"])
return

cos_cache_large_name, sin_cache_large_name = "cos_cache_large", "sin_cache_large"
cos_cache_large, sin_cache_large = self.make_rotary_embedding_caches(rotemb, cos_cache_name=cos_cache_large_name, sin_cache_name=sin_cache_large_name)

Expand Down Expand Up @@ -931,7 +940,6 @@ def make_rotary_embedding_multi_cache(self):
greater_inputs = [f"{gather_name}/output_0", f"/model/constants/TensorProto.INT64/0D/{self.original_context_length}"]
self.make_greater(greater_name, greater_inputs, shape=[])
if_name = f"{basename}/If"
if_cos_cache_output, if_sin_cache_output = "cos_cache", "sin_cache"
self.make_node(
"If", inputs=[f"{greater_name}/output_0"], outputs=[if_cos_cache_output, if_sin_cache_output], name=if_name,
then_branch=self.make_graph(
Expand Down
Loading