Skip to content

Qualcomm AI Engine Direct - Support EPS for cli and fix mv2 to work in QNN2.45 #18002

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

Open
winskuo-quic wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Qualcomm AI Engine Direct - Support EPS for cli and fix mv2 to work in QNN2.45 #18002

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: 7 additions & 3 deletions examples/qualcomm/oss_scripts/mobilevit_v2.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -22,6 +22,7 @@
build_executorch_binary,
get_backend_type,
make_output_dir,
make_quantizer,
parse_skip_delegation_node,
setup_common_args_and_variables,
SimpleADB,
Expand Down Expand Up @@ -90,9 +91,12 @@ def main(args):

pte_filename = "mobilevit_v2_qnn"
backend = get_backend_type(args.backend)
quant_dtype = {
quantizer = {
QnnExecuTorchBackendType.kGpuBackend: None,
QnnExecuTorchBackendType.kHtpBackend: QuantDtype.use_16a8w,
QnnExecuTorchBackendType.kHtpBackend: make_quantizer(
quant_dtype=QuantDtype.use_16a8w,
eps=2**-10,
),
}[backend]
build_executorch_binary(
module,
Expand All @@ -102,7 +106,7 @@ def main(args):
inputs,
skip_node_id_set=skip_node_id_set,
skip_node_op_set=skip_node_op_set,
quant_dtype=quant_dtype,
custom_quantizer=quantizer,
backend=backend,
shared_buffer=args.shared_buffer,
online_prepare=args.online_prepare,
Expand Down
8 changes: 7 additions & 1 deletion examples/qualcomm/util_scripts/cli.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -135,7 +135,6 @@ def __iter__(self):

def quantize(args):
logger = get_logger()

# get corresponding QnnQuantizer
try:
quant_dtype = getattr(QuantDtype, args.config)
Expand All @@ -147,6 +146,7 @@ def quantize(args):
act_observer=act_observer,
backend=get_backend_type(args.backend),
soc_model=args.model,
eps=args.eps,
)
except Exception:
logger.error(
Expand Down Expand Up @@ -450,6 +450,12 @@ def main():
default="htp",
help="Backend to be deployed ('htp'/'gpu' are currently supported).",
)
sub_quantize.add_argument(
"--eps",
help="EPS value for quantizer. Accepts floating‑point literal. E.g., 0.0009765625.",
type=float,
default=None,
)
sub_quantize.set_defaults(callback=quantize)

sub_compile = subparsers.add_parser(
Expand Down
2 changes: 1 addition & 1 deletion examples/qualcomm/utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -371,9 +371,9 @@ def make_quantizer(
act_symmetric=False,
is_qat=False,
submodule_qconfig_list: Optional[List[Tuple[Callable, ModuleQConfig]]] = None,
eps=None,
backend=QnnExecuTorchBackendType.kHtpBackend,
soc_model="SM8750",
eps=None,
):
quantizer = QnnQuantizer(backend=backend, soc_model=getattr(QcomChipset, soc_model))
quantizer.add_custom_quant_annotations(custom_annotations)
Expand Down
Loading