[torchlib] Fix and improve quantization support #1737
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
Tested with: import torch
class M(torch.nn.Module):
def __init__(self):
super().__init__()
self.linear = torch.nn.Linear(5, 10)
def forward(self, x):
return self.linear(x)
example_inputs = (torch.randn(1, 5),)
m = M().eval()
# Step 1. program capture
from torch._export import capture_pre_autograd_graph
pt2e_torch_model = capture_pre_autograd_graph(m, example_inputs)
# Step 2. quantization
from torch.ao.quantization.quantize_pt2e import (
prepare_pt2e,
convert_pt2e,
)
from torch.ao.quantization.quantizer.xnnpack_quantizer import (
XNNPACKQuantizer,
get_symmetric_quantization_config,
)
quantizer = XNNPACKQuantizer().set_global(get_symmetric_quantization_config())
pt2e_torch_model = prepare_pt2e(pt2e_torch_model, quantizer)
# Run the prepared model with sample input data to ensure that internal observers are populated with correct values
pt2e_torch_model(*example_inputs)
# Convert the prepared model to a quantized model
pt2e_torch_model = convert_pt2e(pt2e_torch_model, fold_quantize=False)
program = torch.export.export(pt2e_torch_model, example_inputs)
# we get a model with aten ops
print(program)
# Convert to ONNX
import torch_onnx
torch_onnx.patch_torch(error_report=True)
onnx_program = torch.onnx.export(program, example_inputs, "quantized.textproto") |
ir_version: 9
producer_name: "torch"
producer_version: "2.3.1+cu121"
graph {
node {
output: "val_0"
name: "node_Constant_0"
op_type: "Constant"
attribute {
name: "value"
t {
data_type: 3
raw_data: "E"
}
type: TENSOR
}
metadata_props {
key: "namespace"
value: ": torch.fx.graph_module.GraphModule/quantize_per_tensor: quantized_decomposed.quantize_per_tensor.default"
}
metadata_props {
key: "pkg.torch.onnx.class_hierarchy"
value: "[\'torch.fx.graph_module.GraphModule\', \'quantized_decomposed.quantize_per_tensor.default\']"
}
metadata_props {
key: "pkg.torch.onnx.fx_node"
value: "%quantize_per_tensor : [num_users=1] = call_function[target=torch.ops.quantized_decomposed.quantize_per_tensor.default](args = (%arg2_1, 0.004384273663163185, 69, -128, 127, torch.int8), kwargs = {})"
}
metadata_props {
key: "pkg.torch.onnx.name_scopes"
value: "[\'\', \'quantize_per_tensor\']"
}
metadata_props {
key: "pkg.torch.onnx.stack_trace"
value: " File \"<eval_with_key>.11\", line 7, in forward\n quantize_per_tensor_default = torch.ops.quantized_decomposed.quantize_per_tensor.default(arg0_1, 0.004384273663163185, 69, -128, 127, torch.int8); arg0_1 = None\n"
}
}
node {
output: "val_1"
name: "node_Constant_1"
op_type: "Constant"
attribute {
name: "value"
t {
data_type: 1
raw_data: "\364\251\217;"
}
type: TENSOR
}
metadata_props {
key: "namespace"
value: ": torch.fx.graph_module.GraphModule/quantize_per_tensor: quantized_decomposed.quantize_per_tensor.default"
}
metadata_props {
key: "pkg.torch.onnx.class_hierarchy"
value: "[\'torch.fx.graph_module.GraphModule\', \'quantized_decomposed.quantize_per_tensor.default\']"
}
metadata_props {
key: "pkg.torch.onnx.fx_node"
value: "%quantize_per_tensor : [num_users=1] = call_function[target=torch.ops.quantized_decomposed.quantize_per_tensor.default](args = (%arg2_1, 0.004384273663163185, 69, -128, 127, torch.int8), kwargs = {})"
}
metadata_props {
key: "pkg.torch.onnx.name_scopes"
value: "[\'\', \'quantize_per_tensor\']"
}
metadata_props {
key: "pkg.torch.onnx.stack_trace"
value: " File \"<eval_with_key>.11\", line 7, in forward\n quantize_per_tensor_default = torch.ops.quantized_decomposed.quantize_per_tensor.default(arg0_1, 0.004384273663163185, 69, -128, 127, torch.int8); arg0_1 = None\n"
}
}
node {
input: "arg2_1"
input: "val_1"
input: "val_0"
output: "val_quantize_per_tensor"
name: "node_QuantizeLinear_2"
op_type: "QuantizeLinear"
attribute {
name: "axis"
i: 1
type: INT
}
metadata_props {
key: "namespace"
value: ": torch.fx.graph_module.GraphModule/quantize_per_tensor: quantized_decomposed.quantize_per_tensor.default"
}
metadata_props {
key: "pkg.torch.onnx.class_hierarchy"
value: "[\'torch.fx.graph_module.GraphModule\', \'quantized_decomposed.quantize_per_tensor.default\']"
}
metadata_props {
key: "pkg.torch.onnx.fx_node"
value: "%quantize_per_tensor : [num_users=1] = call_function[target=torch.ops.quantized_decomposed.quantize_per_tensor.default](args = (%arg2_1, 0.004384273663163185, 69, -128, 127, torch.int8), kwargs = {})"
}
metadata_props {
key: "pkg.torch.onnx.name_scopes"
value: "[\'\', \'quantize_per_tensor\']"
}
metadata_props {
key: "pkg.torch.onnx.stack_trace"
value: " File \"<eval_with_key>.11\", line 7, in forward\n quantize_per_tensor_default = torch.ops.quantized_decomposed.quantize_per_tensor.default(arg0_1, 0.004384273663163185, 69, -128, 127, torch.int8); arg0_1 = None\n"
}
}
node {
output: "val_2"
name: "node_Constant_3"
op_type: "Constant"
attribute {
name: "value"
t {
data_type: 3
raw_data: "E"
}
type: TENSOR
}
metadata_props {
key: "namespace"
value: ": torch.fx.graph_module.GraphModule/dequantize_per_tensor: quantized_decomposed.dequantize_per_tensor.default"
}
metadata_props {
key: "pkg.torch.onnx.class_hierarchy"
value: "[\'torch.fx.graph_module.GraphModule\', \'quantized_decomposed.dequantize_per_tensor.default\']"
}
metadata_props {
key: "pkg.torch.onnx.fx_node"
value: "%dequantize_per_tensor : [num_users=1] = call_function[target=torch.ops.quantized_decomposed.dequantize_per_tensor.default](args = (%quantize_per_tensor, 0.004384273663163185, 69, -128, 127, torch.int8), kwargs = {})"
}
metadata_props {
key: "pkg.torch.onnx.name_scopes"
value: "[\'\', \'dequantize_per_tensor\']"
}
metadata_props {
key: "pkg.torch.onnx.stack_trace"
value: " File \"/home/justinchu/dev/torch-onnx/quant.py\", line 10, in forward\n return self.linear(x)\n"
}
}
node {
input: "val_quantize_per_tensor"
input: "val_1"
input: "val_2"
output: "val_dequantize_per_tensor"
name: "node_DequantizeLinear_4"
op_type: "DequantizeLinear"
attribute {
name: "axis"
i: 1
type: INT
}
metadata_props {
key: "namespace"
value: ": torch.fx.graph_module.GraphModule/dequantize_per_tensor: quantized_decomposed.dequantize_per_tensor.default"
}
metadata_props {
key: "pkg.torch.onnx.class_hierarchy"
value: "[\'torch.fx.graph_module.GraphModule\', \'quantized_decomposed.dequantize_per_tensor.default\']"
}
metadata_props {
key: "pkg.torch.onnx.fx_node"
value: "%dequantize_per_tensor : [num_users=1] = call_function[target=torch.ops.quantized_decomposed.dequantize_per_tensor.default](args = (%quantize_per_tensor, 0.004384273663163185, 69, -128, 127, torch.int8), kwargs = {})"
}
metadata_props {
key: "pkg.torch.onnx.name_scopes"
value: "[\'\', \'dequantize_per_tensor\']"
}
metadata_props {
key: "pkg.torch.onnx.stack_trace"
value: " File \"/home/justinchu/dev/torch-onnx/quant.py\", line 10, in forward\n return self.linear(x)\n"
}
}
node {
output: "val_3"
name: "node_Constant_5"
op_type: "Constant"
attribute {
name: "value"
t {
data_type: 3
raw_data: "\000"
}
type: TENSOR
}
metadata_props {
key: "namespace"
value: ": torch.fx.graph_module.GraphModule/quantize_per_tensor_1: quantized_decomposed.quantize_per_tensor.default"
}
metadata_props {
key: "pkg.torch.onnx.class_hierarchy"
value: "[\'torch.fx.graph_module.GraphModule\', \'quantized_decomposed.quantize_per_tensor.default\']"
}
metadata_props {
key: "pkg.torch.onnx.fx_node"
value: "%quantize_per_tensor_1 : [num_users=1] = call_function[target=torch.ops.quantized_decomposed.quantize_per_tensor.default](args = (%arg0_1, 0.003469259710982442, 0, -127, 127, torch.int8), kwargs = {})"
}
metadata_props {
key: "pkg.torch.onnx.name_scopes"
value: "[\'\', \'quantize_per_tensor_1\']"
}
metadata_props {
key: "pkg.torch.onnx.stack_trace"
value: " File \"/home/justinchu/dev/torch-onnx/quant.py\", line 10, in forward\n return self.linear(x)\n"
}
}
node {
output: "val_4"
name: "node_Constant_6"
op_type: "Constant"
attribute {
name: "value"
t {
data_type: 1
raw_data: "\205\\c;"
}
type: TENSOR
}
metadata_props {
key: "namespace"
value: ": torch.fx.graph_module.GraphModule/quantize_per_tensor_1: quantized_decomposed.quantize_per_tensor.default"
}
metadata_props {
key: "pkg.torch.onnx.class_hierarchy"
value: "[\'torch.fx.graph_module.GraphModule\', \'quantized_decomposed.quantize_per_tensor.default\']"
}
metadata_props {
key: "pkg.torch.onnx.fx_node"
value: "%quantize_per_tensor_1 : [num_users=1] = call_function[target=torch.ops.quantized_decomposed.quantize_per_tensor.default](args = (%arg0_1, 0.003469259710982442, 0, -127, 127, torch.int8), kwargs = {})"
}
metadata_props {
key: "pkg.torch.onnx.name_scopes"
value: "[\'\', \'quantize_per_tensor_1\']"
}
metadata_props {
key: "pkg.torch.onnx.stack_trace"
value: " File \"/home/justinchu/dev/torch-onnx/quant.py\", line 10, in forward\n return self.linear(x)\n"
}
}
node {
input: "linear_weight"
input: "val_4"
input: "val_3"
output: "val_quantize_per_tensor_1"
name: "node_QuantizeLinear_7"
op_type: "QuantizeLinear"
attribute {
name: "axis"
i: 1
type: INT
}
metadata_props {
key: "namespace"
value: ": torch.fx.graph_module.GraphModule/quantize_per_tensor_1: quantized_decomposed.quantize_per_tensor.default"
}
metadata_props {
key: "pkg.torch.onnx.class_hierarchy"
value: "[\'torch.fx.graph_module.GraphModule\', \'quantized_decomposed.quantize_per_tensor.default\']"
}
metadata_props {
key: "pkg.torch.onnx.fx_node"
value: "%quantize_per_tensor_1 : [num_users=1] = call_function[target=torch.ops.quantized_decomposed.quantize_per_tensor.default](args = (%arg0_1, 0.003469259710982442, 0, -127, 127, torch.int8), kwargs = {})"
}
metadata_props {
key: "pkg.torch.onnx.name_scopes"
value: "[\'\', \'quantize_per_tensor_1\']"
}
metadata_props {
key: "pkg.torch.onnx.stack_trace"
value: " File \"/home/justinchu/dev/torch-onnx/quant.py\", line 10, in forward\n return self.linear(x)\n"
}
}
node {
output: "val_5"
name: "node_Constant_8"
op_type: "Constant"
attribute {
name: "value"
t {
data_type: 3
raw_data: "\000"
}
type: TENSOR
}
metadata_props {
key: "namespace"
value: ": torch.fx.graph_module.GraphModule/dequantize_per_tensor_1: quantized_decomposed.dequantize_per_tensor.default"
}
metadata_props {
key: "pkg.torch.onnx.class_hierarchy"
value: "[\'torch.fx.graph_module.GraphModule\', \'quantized_decomposed.dequantize_per_tensor.default\']"
}
metadata_props {
key: "pkg.torch.onnx.fx_node"
value: "%dequantize_per_tensor_1 : [num_users=1] = call_function[target=torch.ops.quantized_decomposed.dequantize_per_tensor.default](args = (%quantize_per_tensor_1, 0.003469259710982442, 0, -127, 127, torch.int8), kwargs = {})"
}
metadata_props {
key: "pkg.torch.onnx.name_scopes"
value: "[\'\', \'dequantize_per_tensor_1\']"
}
metadata_props {
key: "pkg.torch.onnx.stack_trace"
value: " File \"/home/justinchu/dev/torch-onnx/quant.py\", line 10, in forward\n return self.linear(x)\n"
}
}
node {
input: "val_quantize_per_tensor_1"
input: "val_4"
input: "val_5"
output: "val_dequantize_per_tensor_1"
name: "node_DequantizeLinear_9"
op_type: "DequantizeLinear"
attribute {
name: "axis"
i: 1
type: INT
}
metadata_props {
key: "namespace"
value: ": torch.fx.graph_module.GraphModule/dequantize_per_tensor_1: quantized_decomposed.dequantize_per_tensor.default"
}
metadata_props {
key: "pkg.torch.onnx.class_hierarchy"
value: "[\'torch.fx.graph_module.GraphModule\', \'quantized_decomposed.dequantize_per_tensor.default\']"
}
metadata_props {
key: "pkg.torch.onnx.fx_node"
value: "%dequantize_per_tensor_1 : [num_users=1] = call_function[target=torch.ops.quantized_decomposed.dequantize_per_tensor.default](args = (%quantize_per_tensor_1, 0.003469259710982442, 0, -127, 127, torch.int8), kwargs = {})"
}
metadata_props {
key: "pkg.torch.onnx.name_scopes"
value: "[\'\', \'dequantize_per_tensor_1\']"
}
metadata_props {
key: "pkg.torch.onnx.stack_trace"
value: " File \"/home/justinchu/dev/torch-onnx/quant.py\", line 10, in forward\n return self.linear(x)\n"
}
}
node {
input: "val_dequantize_per_tensor_1"
output: "val_t"
name: "node_Transpose_10"
op_type: "Transpose"
attribute {
name: "perm"
ints: 1
ints: 0
type: INTS
}
metadata_props {
key: "namespace"
value: ": torch.fx.graph_module.GraphModule/linear: torch.nn.modules.linear.Linear/t: aten.t.default"
}
metadata_props {
key: "pkg.torch.onnx.class_hierarchy"
value: "[\'torch.fx.graph_module.GraphModule\', \'torch.nn.modules.linear.Linear\', \'aten.t.default\']"
}
metadata_props {
key: "pkg.torch.onnx.fx_node"
value: "%t : [num_users=1] = call_function[target=torch.ops.aten.t.default](args = (%dequantize_per_tensor_1,), kwargs = {})"
}
metadata_props {
key: "pkg.torch.onnx.name_scopes"
value: "[\'\', \'linear\', \'t\']"
}
metadata_props {
key: "pkg.torch.onnx.stack_trace"
value: " File \"/home/justinchu/dev/torch-onnx/quant.py\", line 10, in forward\n return self.linear(x)\n"
}
}
node {
input: "linear_bias"
input: "val_dequantize_per_tensor"
input: "val_t"
output: "val_addmm"
name: "node_aten_addmm_11"
op_type: "aten_addmm"
attribute {
name: "beta"
f: 1.0
type: FLOAT
}
attribute {
name: "alpha"
f: 1.0
type: FLOAT
}
domain: "pkg.onnxscript.torch_lib"
metadata_props {
key: "namespace"
value: ": torch.fx.graph_module.GraphModule/linear: torch.nn.modules.linear.Linear/addmm: aten.addmm.default"
}
metadata_props {
key: "pkg.torch.onnx.class_hierarchy"
value: "[\'torch.fx.graph_module.GraphModule\', \'torch.nn.modules.linear.Linear\', \'aten.addmm.default\']"
}
metadata_props {
key: "pkg.torch.onnx.fx_node"
value: "%addmm : [num_users=1] = call_function[target=torch.ops.aten.addmm.default](args = (%arg1_1, %dequantize_per_tensor, %t), kwargs = {})"
}
metadata_props {
key: "pkg.torch.onnx.name_scopes"
value: "[\'\', \'linear\', \'addmm\']"
}
metadata_props {
key: "pkg.torch.onnx.stack_trace"
value: " File \"/home/justinchu/dev/torch-onnx/quant.py\", line 10, in forward\n return self.linear(x)\n"
}
}
node {
output: "val_6"
name: "node_Constant_12"
op_type: "Constant"
attribute {
name: "value"
t {
data_type: 3
raw_data: "\314"
}
type: TENSOR
}
metadata_props {
key: "namespace"
value: ": torch.fx.graph_module.GraphModule/quantize_per_tensor_2: quantized_decomposed.quantize_per_tensor.default"
}
metadata_props {
key: "pkg.torch.onnx.class_hierarchy"
value: "[\'torch.fx.graph_module.GraphModule\', \'quantized_decomposed.quantize_per_tensor.default\']"
}
metadata_props {
key: "pkg.torch.onnx.fx_node"
value: "%quantize_per_tensor_2 : [num_users=1] = call_function[target=torch.ops.quantized_decomposed.quantize_per_tensor.default](args = (%addmm, 0.003849889151751995, -52, -128, 127, torch.int8), kwargs = {})"
}
metadata_props {
key: "pkg.torch.onnx.name_scopes"
value: "[\'\', \'quantize_per_tensor_2\']"
}
metadata_props {
key: "pkg.torch.onnx.stack_trace"
value: " File \"/home/justinchu/dev/torch-onnx/quant.py\", line 10, in forward\n return self.linear(x)\n"
}
}
node {
output: "val_7"
name: "node_Constant_13"
op_type: "Constant"
attribute {
name: "value"
t {
data_type: 1
raw_data: "lN|;"
}
type: TENSOR
}
metadata_props {
key: "namespace"
value: ": torch.fx.graph_module.GraphModule/quantize_per_tensor_2: quantized_decomposed.quantize_per_tensor.default"
}
metadata_props {
key: "pkg.torch.onnx.class_hierarchy"
value: "[\'torch.fx.graph_module.GraphModule\', \'quantized_decomposed.quantize_per_tensor.default\']"
}
metadata_props {
key: "pkg.torch.onnx.fx_node"
value: "%quantize_per_tensor_2 : [num_users=1] = call_function[target=torch.ops.quantized_decomposed.quantize_per_tensor.default](args = (%addmm, 0.003849889151751995, -52, -128, 127, torch.int8), kwargs = {})"
}
metadata_props {
key: "pkg.torch.onnx.name_scopes"
value: "[\'\', \'quantize_per_tensor_2\']"
}
metadata_props {
key: "pkg.torch.onnx.stack_trace"
value: " File \"/home/justinchu/dev/torch-onnx/quant.py\", line 10, in forward\n return self.linear(x)\n"
}
}
node {
input: "val_addmm"
input: "val_7"
input: "val_6"
output: "val_quantize_per_tensor_2"
name: "node_QuantizeLinear_14"
op_type: "QuantizeLinear"
attribute {
name: "axis"
i: 1
type: INT
}
metadata_props {
key: "namespace"
value: ": torch.fx.graph_module.GraphModule/quantize_per_tensor_2: quantized_decomposed.quantize_per_tensor.default"
}
metadata_props {
key: "pkg.torch.onnx.class_hierarchy"
value: "[\'torch.fx.graph_module.GraphModule\', \'quantized_decomposed.quantize_per_tensor.default\']"
}
metadata_props {
key: "pkg.torch.onnx.fx_node"
value: "%quantize_per_tensor_2 : [num_users=1] = call_function[target=torch.ops.quantized_decomposed.quantize_per_tensor.default](args = (%addmm, 0.003849889151751995, -52, -128, 127, torch.int8), kwargs = {})"
}
metadata_props {
key: "pkg.torch.onnx.name_scopes"
value: "[\'\', \'quantize_per_tensor_2\']"
}
metadata_props {
key: "pkg.torch.onnx.stack_trace"
value: " File \"/home/justinchu/dev/torch-onnx/quant.py\", line 10, in forward\n return self.linear(x)\n"
}
}
node {
output: "val_8"
name: "node_Constant_15"
op_type: "Constant"
attribute {
name: "value"
t {
data_type: 3
raw_data: "\314"
}
type: TENSOR
}
metadata_props {
key: "namespace"
value: ": torch.fx.graph_module.GraphModule/dequantize_per_tensor_2: quantized_decomposed.dequantize_per_tensor.default"
}
metadata_props {
key: "pkg.torch.onnx.class_hierarchy"
value: "[\'torch.fx.graph_module.GraphModule\', \'quantized_decomposed.dequantize_per_tensor.default\']"
}
metadata_props {
key: "pkg.torch.onnx.fx_node"
value: "%dequantize_per_tensor_2 : [num_users=1] = call_function[target=torch.ops.quantized_decomposed.dequantize_per_tensor.default](args = (%quantize_per_tensor_2, 0.003849889151751995, -52, -128, 127, torch.int8), kwargs = {})"
}
metadata_props {
key: "pkg.torch.onnx.name_scopes"
value: "[\'\', \'dequantize_per_tensor_2\']"
}
metadata_props {
key: "pkg.torch.onnx.stack_trace"
value: " File \"<eval_with_key>.11\", line 15, in forward\n dequantize_per_tensor_default_2 = torch.ops.quantized_decomposed.dequantize_per_tensor.default(quantize_per_tensor_default_2, 0.003849889151751995, -52, -128, 127, torch.int8); quantize_per_tensor_default_2 = None\n"
}
}
node {
input: "val_quantize_per_tensor_2"
input: "val_7"
input: "val_8"
output: "val_dequantize_per_tensor_2"
name: "node_DequantizeLinear_16"
op_type: "DequantizeLinear"
attribute {
name: "axis"
i: 1
type: INT
}
metadata_props {
key: "namespace"
value: ": torch.fx.graph_module.GraphModule/dequantize_per_tensor_2: quantized_decomposed.dequantize_per_tensor.default"
}
metadata_props {
key: "pkg.torch.onnx.class_hierarchy"
value: "[\'torch.fx.graph_module.GraphModule\', \'quantized_decomposed.dequantize_per_tensor.default\']"
}
metadata_props {
key: "pkg.torch.onnx.fx_node"
value: "%dequantize_per_tensor_2 : [num_users=1] = call_function[target=torch.ops.quantized_decomposed.dequantize_per_tensor.default](args = (%quantize_per_tensor_2, 0.003849889151751995, -52, -128, 127, torch.int8), kwargs = {})"
}
metadata_props {
key: "pkg.torch.onnx.name_scopes"
value: "[\'\', \'dequantize_per_tensor_2\']"
}
metadata_props {
key: "pkg.torch.onnx.stack_trace"
value: " File \"<eval_with_key>.11\", line 15, in forward\n dequantize_per_tensor_default_2 = torch.ops.quantized_decomposed.dequantize_per_tensor.default(quantize_per_tensor_default_2, 0.003849889151751995, -52, -128, 127, torch.int8); quantize_per_tensor_default_2 = None\n"
}
}
name: "main_graph"
initializer {
dims: 10
dims: 5
data_type: 1
name: "linear_weight"
raw_data: "*\372\310\275\245\232\206=\314\225\341\276\371\214}\276\374%7\276\262\354\313<\205\241\273\276Q1\266>L\272v>N\rf>\200\263e>\321W\336>\017\016M\276\205\204\014\274\361@\253>\351\372\246\276\201\r\271\275\3015~\276\300\010N\276;\205n\276\002U\216=\033\253\340>l\016\031>\035\264\030>\311\032\226>\345w]\276*\346\262\275\300\356\025\276\034\201\271\276~~e>\010\245\310>\204\206\336\275\202,\\>%\3754\276\271\360\310>\242@\272>O.\226\275D\340\271\276s~\313\276~*\231\276\002*\276\276\360\032\222>\"\262G\276\227\326\217=GVC>\225\r\234>v\370,=\314\223\206\276\225\212\223>\337\002Q\275"
}
initializer {
dims: 10
data_type: 1
name: "linear_bias"
raw_data: "V\336\021\276TC4>\306y\250>h\225\277>\330\351\326>\'5n\274\275\036\270>\341bf\276\366\230\202\2768\222\303>"
}
input {
name: "arg2_1"
type {
tensor_type {
elem_type: 1
shape {
dim {
dim_value: 1
}
dim {
dim_value: 5
}
}
}
}
metadata_props {
key: "pkg.torch.export.graph_signature.InputSpec.kind"
value: "USER_INPUT"
}
metadata_props {
key: "pkg.torch.export.graph_signature.InputSpec.persistent"
value: "None"
}
}
input {
name: "linear_weight"
type {
tensor_type {
elem_type: 1
shape {
dim {
dim_value: 10
}
dim {
dim_value: 5
}
}
}
}
metadata_props {
key: "pkg.torch.export.graph_signature.InputSpec.kind"
value: "PARAMETER"
}
metadata_props {
key: "pkg.torch.export.graph_signature.InputSpec.persistent"
value: "None"
}
metadata_props {
key: "pkg.torch.onnx.original_node_name"
value: "arg0_1"
}
}
input {
name: "linear_bias"
type {
tensor_type {
elem_type: 1
shape {
dim {
dim_value: 10
}
}
}
}
metadata_props {
key: "pkg.torch.export.graph_signature.InputSpec.kind"
value: "PARAMETER"
}
metadata_props {
key: "pkg.torch.export.graph_signature.InputSpec.persistent"
value: "None"
}
metadata_props {
key: "pkg.torch.onnx.original_node_name"
value: "arg1_1"
}
}
output {
name: "val_dequantize_per_tensor_2"
type {
tensor_type {
elem_type: 1
shape {
dim {
dim_value: 1
}
dim {
dim_value: 10
}
}
}
}
metadata_props {
key: "pkg.torch.export.graph_signature.OutputSpec.kind"
value: "USER_OUTPUT"
}
}
value_info {
name: "val_quantize_per_tensor"
type {
tensor_type {
elem_type: 3
shape {
dim {
dim_value: 1
}
dim {
dim_value: 5
}
}
}
}
}
value_info {
name: "val_dequantize_per_tensor"
type {
tensor_type {
elem_type: 1
shape {
dim {
dim_value: 1
}
dim {
dim_value: 5
}
}
}
}
}
value_info {
name: "val_quantize_per_tensor_1"
type {
tensor_type {
elem_type: 3
shape {
dim {
dim_value: 10
}
dim {
dim_value: 5
}
}
}
}
}
value_info {
name: "val_dequantize_per_tensor_1"
type {
tensor_type {
elem_type: 1
shape {
dim {
dim_value: 10
}
dim {
dim_value: 5
}
}
}
}
}
value_info {
name: "val_t"
type {
tensor_type {
elem_type: 1
shape {
dim {
dim_value: 5
}
dim {
dim_value: 10
}
}
}
}
}
value_info {
name: "val_addmm"
type {
tensor_type {
elem_type: 1
shape {
dim {
dim_value: 1
}
dim {
dim_value: 10
}
}
}
}
}
value_info {
name: "val_quantize_per_tensor_2"
type {
tensor_type {
elem_type: 3
shape {
dim {
dim_value: 1
}
dim {
dim_value: 10
}
}
}
}
}
}
opset_import {
domain: ""
version: 18
}
opset_import {
domain: "pkg.onnxscript.torch_lib"
version: 1
}
opset_import {
domain: "pkg.onnxscript.torch_lib.common"
version: 1
}
functions {
name: "aten_addmm"
input: "self"
input: "mat1"
input: "mat2"
output: "return_val"
node {
input: "mat1"
input: "mat2"
input: "self"
output: "return_val"
name: "n0"
op_type: "Gemm"
attribute {
name: "alpha"
type: FLOAT
ref_attr_name: "alpha"
}
attribute {
name: "beta"
type: FLOAT
ref_attr_name: "beta"
}
}
doc_string: "addmm(Tensor self, Tensor mat1, Tensor mat2, *, Scalar beta=1, Scalar alpha=1) -> Tensor"
opset_import {
domain: ""
version: 18
}
domain: "pkg.onnxscript.torch_lib"
attribute_proto {
name: "beta"
f: 1.0
type: FLOAT
}
attribute_proto {
name: "alpha"
f: 1.0
type: FLOAT
}
} |
justinchuby
requested review from
gramalingam,
titaiwangms,
xadupre and
shubhambhokare1
justinchuby
added
the
topic: torch_lib
Codecov ReportAttention: Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #1737 +/- ##
==========================================
- Coverage 74.89% 74.73% -0.17%
==========================================
Files 245 245
Lines 26358 26369 +11
Branches 4793 4795 +2
==========================================
- Hits 19742 19706 -36
- Misses 5695 5743 +48
+ Partials 921 920 -1 ☔ View full report in Codecov by Sentry. |
titaiwangms
approved these changes
Jul 18, 2024
![github-advanced-security[bot]](https://avatars.githubusercontent.com/in/57789?s=60&v=4){kind=small}
| @@ -56,3 +62,10 @@ | |||
| result = op.Cast(a, to=dtype) | |||
|
|
|||
| return result | |||
|
|
|||
|
|
|||
| def constant(array, dtype: int | onnx.TensorProto.DataType | ir.DataType) -> TensorType: | |||
Check failure
Code scanning / lintrunner
PYLINT/E0601 Error
Test Results 24 files ± 0 24 suites ±0 1h 51m 44s ⏱️ + 3m 53s For more details on these failures, see this check. Results for commit 64e6416. ± Comparison against base commit 2401de4. ♻️ This comment has been updated with latest results. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Fix bugs in the implementation where the dtype of the zero point is not correctly set. Tested with exporter.