Skip to content

Arm backend: Add Qwen3 VL language model MXFP8 test #20533

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
tom-arm wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Arm backend: Add Qwen3 VL language model MXFP8 test #20533

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
39 changes: 39 additions & 0 deletions backends/arm/test/models/Qwen3_VL/test_qwen3_vl_model.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -11,10 +11,12 @@
import pytest
import torch
import torch.nn.functional as F
from executorch.backends.arm.ao_ext import MXFPOpConfig
from executorch.backends.arm.test import common
from executorch.backends.arm.test.models.Qwen3_VL.qwen3_vl_test_config import (
get_qwen3_vl_2b_instruct_checkpoint_config,
)
from executorch.backends.arm.test.ops.mxfp.common import MXFPTosaPipelineFP
from executorch.backends.arm.test.tester.test_pipeline import (
TosaPipelineFP,
VgfPipeline,
Expand All @@ -25,6 +27,7 @@
)

input_t = Tuple[torch.Tensor, ...]
aten_op_mxfp_linear = "torch.ops.tosa_mxfp.linear.default"


def _make_qwen3_vl_2b_instruct_layer_config():
Expand Down Expand Up @@ -104,6 +107,10 @@ def _to_bfloat16_model_and_floating_inputs(
)


def _is_linear(module: torch.nn.Module, _fqn: str) -> bool:
return isinstance(module, torch.nn.Linear)


class TextModelWrapper(Qwen3VLModelTestModule):
def __init__(self, config) -> None:
super().__init__()
Expand Down Expand Up @@ -300,3 +307,35 @@ def test_qwen3_vl_full_models_vgf_no_quant_bf16(test_case: Qwen3VLModelTestCase)
tosa_spec="TOSA-1.0+FP+bf16",
)
pipeline.run()


@pytest.mark.slow
def test_qwen3_vl_text_model_tosa_mxfp8_bf16():
# The Qwen 3 VL FP8 model only quantizes the TextModel
model, inputs = TextModelWrapper.prepare_model_and_inputs()
model, inputs = _to_bfloat16_model_and_floating_inputs(model, inputs)
mxfp_config = MXFPOpConfig(weight_dtype=torch.float8_e4m3fn)
with torch.no_grad():
pipeline = MXFPTosaPipelineFP[input_t](
model,
inputs,
aten_op=aten_op_mxfp_linear,
exir_op=[],
filter_fn=_is_linear,
frobenius_threshold=0.1,
cosine_threshold=0.98,
mxfp_config=mxfp_config,
tosa_version="1.1",
tosa_extensions=["bf16", "mxfp"],
)
# Check all linear layers are converted to MXFP
linear_count = sum(
_is_linear(submodule, name) for name, submodule in model.named_modules()
)
pipeline.add_stage_after(
"export",
pipeline.tester.check_count,
{aten_op_mxfp_linear: linear_count},
suffix="mxfp_linear",
)
pipeline.run()
Loading