Skip to content

Updates to use torchao's updated choose_qparams_affine and quantize/dequantize_affine #11070

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
merged 1 commit into from
May 24, 2025
Merged

Updates to use torchao's updated choose_qparams_affine and quantize/dequantize_affine #11070

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
3 changes: 0 additions & 3 deletions backends/xnnpack/utils/quant_utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -222,9 +222,6 @@ def extract_qdq_affine_op_args_for_decomposed_ops(node: torch.fx.Node):

# add target_dtype_node after quant_min/quant_max
args.append(target_dtype)
# zero_point_domain
if len(node.args) > 7 and node.args[7] != "INT":
return None, None

if is_per_channel_group(node):
block_sizes = cast(list[int], node.args[1])
Expand Down
3 changes: 0 additions & 3 deletions exir/passes/_quant_patterns_and_replacements.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1017,7 +1017,6 @@ def embedding_byte_dtype_pattern(
torch.int8,
-128,
127,
"INT",
output_dtype,
)
return torch.ops.aten.embedding.default(dq, indices)
Expand Down Expand Up @@ -1062,7 +1061,6 @@ def embedding_2bit_dtype_pattern(
torch.int8,
-2,
1,
"INT",
output_dtype,
)
return torch.ops.aten.embedding.default(dq, indices)
Expand Down Expand Up @@ -1110,7 +1108,6 @@ def embedding_4bit_dtype_pattern(
torch.int8,
-8,
7,
"INT",
output_dtype,
)
return torch.ops.aten.embedding.default(dq, indices)
Expand Down
2 changes: 1 addition & 1 deletion third-party/ao
Open in desktop
Submodule ao updated 26 files
+25 −0 setup.py
+1 −8 test/dtypes/test_uintx.py
+10 −0 test/float8/test_base.py
+223 −0 test/prototype/inductor/test_int8_sdpa_fusion.py
+27 −163 test/quantization/test_quant_primitives.py
+0 −6 test/sparsity/test_marlin.py
+140 −1 test/test_ops.py
+1 −0 torchao/core/config.py
+1,910 −0 torchao/csrc/cpu/int8_sdpa.cpp
+185 −73 torchao/dtypes/affine_quantized_tensor.py
+14 −3 torchao/dtypes/affine_quantized_tensor_ops.py
+6 −8 torchao/dtypes/uintx/int4_cpu_layout.py
+6 −8 torchao/dtypes/uintx/tensor_core_tiled_layout.py
+1 −2 torchao/experimental/quant_passes.py
+89 −0 torchao/ops.py
+0 −0 torchao/prototype/inductor/__init__.py
+35 −0 torchao/prototype/inductor/fx_passes/README.md
+5 −0 torchao/prototype/inductor/fx_passes/__init__.py
+392 −0 torchao/prototype/inductor/fx_passes/int8_sdpa_fusion.py
+28 −9 torchao/prototype/parq/quant/uniform_torchao.py
+4 −0 torchao/quantization/__init__.py
+0 −4 torchao/quantization/pt2e/observer.py
+38 −13 torchao/quantization/qat/affine_fake_quantized_tensor.py
+8 −7 torchao/quantization/quant_api.py
+645 −206 torchao/quantization/quant_primitives.py
+59 −17 torchao/quantization/utils.py
Loading