Skip to content

Qualcomm AI Engine Direct - Improve GA Qwen 2.5 #14047

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 1 commit into from
Sep 9, 2025
Merged

Qualcomm AI Engine Direct - Improve GA Qwen 2.5 #14047

Show file tree
Hide file tree
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
4 changes: 2 additions & 2 deletions backends/qualcomm/tests/test_qnn_delegate.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4979,9 +4979,9 @@ def test_static_qwen2_5(self):
if "Error" in msg:
self.fail(msg["Error"])
else:
inference_speed_ref = {"SM8650": 110, "SM8750": 130}
inference_speed_ref = {"SM8650": 115, "SM8750": 155}
self.assertLessEqual(msg["wiki_ppl"], 15)
self.assertLessEqual(msg["pte_size"], 800000000) # 800mb
self.assertLessEqual(msg["pte_size"], 600000000) # 600mb
if self.model in inference_speed_ref:
self.assertGreaterEqual(
msg["inference_speed"], inference_speed_ref[self.model]
Expand Down
10 changes: 5 additions & 5 deletions examples/qualcomm/oss_scripts/llama/__init__.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -211,8 +211,8 @@ class Qwen2_5_0_5B(LLMModelConfig):

num_sharding = 1
# quant config
ptq = QuantDtype.use_16a8w
group_size = None
ptq = QuantDtype.use_16a4w_block
group_size = 16
masked_softmax = True
r1 = False
r2 = False
Expand All @@ -233,13 +233,13 @@ class Qwen2_5_1_5B(LLMModelConfig):

num_sharding = 1
# quant config
ptq = QuantDtype.use_16a8w
group_size = None
ptq = QuantDtype.use_16a4w_block
group_size = 16
masked_softmax = True
r1 = False
r2 = False
r3 = True
custom_annotation = ()
custom_annotation = (annotate_output_16a8w,)


@register_llm_model("qwen3-0_6b")
Expand Down
Loading