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 default int8 quantization for CLI #592

Merged
merged 1 commit into from
Mar 7, 2024
Merged

Fix default int8 quantization for CLI #592

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
19 changes: 9 additions & 10 deletions optimum/commands/export/openvino.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -157,13 +157,12 @@ def run(self):
)
self.args.weight_format = "int8"

weight_format = self.args.weight_format or "fp32"

ov_config = None
if weight_format in {"fp16", "fp32"}:
ov_config = OVConfig(dtype=weight_format)
if self.args.weight_format is None:
ov_config = None
elif self.args.weight_format in {"fp16", "fp32"}:
ov_config = OVConfig(dtype=self.args.weight_format)
else:
is_int8 = weight_format == "int8"
is_int8 = self.args.weight_format == "int8"

# For int4 quantization if not parameter is provided, then use the default config if exist
if (
Expand All @@ -182,12 +181,12 @@ def run(self):
"group_size": -1 if is_int8 else self.args.group_size,
}

if weight_format in {"int4_sym_g128", "int4_asym_g128", "int4_sym_g64", "int4_asym_g64"}:
if self.args.weight_format in {"int4_sym_g128", "int4_asym_g128", "int4_sym_g64", "int4_asym_g64"}:
logger.warning(
f"--weight-format {weight_format} is deprecated, possible choices are fp32, fp16, int8, int4"
f"--weight-format {self.args.weight_format} is deprecated, possible choices are fp32, fp16, int8, int4"
)
quantization_config["sym"] = "asym" not in weight_format
quantization_config["group_size"] = 128 if "128" in weight_format else 64
quantization_config["sym"] = "asym" not in self.args.weight_format
quantization_config["group_size"] = 128 if "128" in self.args.weight_format else 64
ov_config = OVConfig(quantization_config=quantization_config)

# TODO : add input shapes
Expand Down
17 changes: 2 additions & 15 deletions optimum/exporters/openvino/__main__.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -21,26 +21,13 @@

from optimum.exporters import TasksManager
from optimum.exporters.onnx.base import OnnxConfig
from optimum.exporters.onnx.constants import SDPA_ARCHS_ONNX_EXPORT_NOT_SUPPORTED
from optimum.utils.save_utils import maybe_load_preprocessors

from ...intel.utils.import_utils import (
is_openvino_tokenizers_available,
is_optimum_version,
is_transformers_version,
)
from ...intel.utils.import_utils import is_openvino_tokenizers_available, is_transformers_version
from .convert import export_from_model, export_tokenizer


if is_optimum_version(">=", "1.16.0"):
from optimum.exporters.onnx.constants import SDPA_ARCHS_ONNX_EXPORT_NOT_SUPPORTED
else:
# Copied from https://github.com/huggingface/optimum/blob/main/optimum/exporters/onnx/constants.py
SDPA_ARCHS_ONNX_EXPORT_NOT_SUPPORTED = [
"bart",
"whisper",
]


if TYPE_CHECKING:
from optimum.intel.openvino.configuration import OVConfig

Expand Down
Loading