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

disable kv cache compression for fp vlm #1080

Merged
merged 1 commit into from
Dec 19, 2024
Merged

disable kv cache compression for fp vlm #1080

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
10 changes: 8 additions & 2 deletions optimum/exporters/openvino/convert.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -100,9 +100,15 @@ def _set_runtime_options(
for model_name in models_and_export_configs.keys():
_, sub_export_config = models_and_export_configs[model_name]
sub_export_config.runtime_options = {}
if "diffusers" in library_name or "text-generation" in task:
if (
"diffusers" in library_name
or "text-generation" in task
or ("image-text-to-text" in task and model_name == "language_model")
):
sub_export_config.runtime_options["ACTIVATIONS_SCALE_FACTOR"] = "8.0"
if not quantized_model and "text-generation" in task:
if not quantized_model and (
"text-generation" in task or ("image-text-to-text" in task and model_name == "language_model")
):
sub_export_config.runtime_options["KV_CACHE_PRECISION"] = "f16"


Expand Down
8 changes: 8 additions & 0 deletions tests/openvino/test_export.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -134,6 +134,14 @@ def _openvino_export(
self.assertTrue(ov_model.model.has_rt_info(["runtime_options", "ACTIVATIONS_SCALE_FACTOR"]))
self.assertTrue(ov_model.model.has_rt_info(["runtime_options", "KV_CACHE_PRECISION"]))

if task == "image-text-to-text":
self.assertTrue(
ov_model.language_model.model.has_rt_info(["runtime_options", "KV_CACHE_PRECISION"])
)
self.assertTrue(
ov_model.language_model.model.has_rt_info(["runtime_options", "ACTIVATIONS_SCALE_FACTOR"])
)

if library_name == "diffusers":
self.assertTrue(
ov_model.vae_encoder.model.has_rt_info(["runtime_options", "ACTIVATIONS_SCALE_FACTOR"])
Expand Down
Loading