Skip to content

[Relay][FQ2I]: Use appropriate dtype while quantizing relay.op.nn.pad… #17177

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
quic-sanirudh merged 2 commits into from
Jul 22, 2024
Merged

[Relay][FQ2I]: Use appropriate dtype while quantizing relay.op.nn.pad… #17177

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
ee654af
[Relay][FQ2I]: Use appropriate dtype while quantizing relay.op.nn.pad…
arangasa Jul 19, 2024
54d81c8
Keep default axis
arangasa Jul 20, 2024
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
2 changes: 1 addition & 1 deletion python/tvm/relay/transform/fake_quantization_to_integer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -466,7 +466,7 @@ def pad(expr, type_map):
# If the pad-value is a constant, we need to quantize it
assert isinstance(pad_value, relay.expr.Constant)
assert pad_value.checked_type.dtype in ["float32", "float64", "float16", "bfloat16"]
pad_value = relay.qnn.op.quantize(pad_value, t.scale, t.zero_point)
pad_value = relay.qnn.op.quantize(pad_value, t.scale, t.zero_point, out_dtype=t.dtype)

out = relay.op.nn.pad(arg, pad_value=pad_value, **expr.attrs)
return [out, t]
Expand Down
14 changes: 14 additions & 0 deletions tests/python/relay/test_pass_fake_quantization_to_integer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -814,6 +814,20 @@ def test_fake_quantize_pad():
compare_fq_to_int(op, [x_np])


def test_fake_quantize_pad_with_float_min():
in_shape = [1, 383, 128]
x = relay.var("x", shape=in_shape, dtype="float32")
op = relay.qnn.quantize(x, relay.const(1.0), relay.const(0), out_dtype="uint8")
op = relay.qnn.dequantize(op, relay.const(1.0), relay.const(0), out_dtype="float32")
op = relay.op.nn.pad(
op, pad_width=[[0, 0], [0, 1], [0, 0]], pad_value=relay.const(-3.40282e38, dtype="float32")
)
op = relay.qnn.op.quantize(op, relay.const(1.0), relay.const(0), out_dtype="uint8")
x_np = np.random.randint(0, 256, size=in_shape)
x_as_float = x_np.astype("float32")
compare_fq_to_int(op, [x_as_float], True)


def test_fake_quantize_depth_to_space():
x = relay.var("x", shape=[1, 3, 224, 224], dtype="int8")

Expand Down