torch._dynamo.exc.Unsupported: Unsupported: quantized nyi in meta tensors with fake tensor propagation.

[gpchowdari]

🐛 Describe the bug

to_edge_transform_and_lower throwing the error when quantized input is passed in export function.

Sample program to reproduce:

import torch

from torch.export import export_for_training
from executorch.exir import EdgeCompileConfig, to_edge_transform_and_lower
from torch.ao.quantization.quantize_pt2e import convert_pt2e, prepare_pt2e

from executorch.backends.xnnpack.quantizer.xnnpack_quantizer import (
    get_symmetric_quantization_config,
    XNNPACKQuantizer,
)
from executorch.backends.xnnpack.partition.xnnpack_partitioner import XnnpackPartitioner

from torch.export.dynamic_shapes import Dim
from torch.export import export, ExportedProgram
from executorch.backends.xnnpack.utils.configs import get_xnnpack_edge_compile_config


class MyModule(torch.nn.Module):
    def __init__(self):
        super().__init__()
        self.fc = torch.nn.Linear(50,10)
        
    
    def forward(self, x,ilens):
        xd = self.fc(x)
        il = ilens
        return xd,il

def xnnpack_quantize_n_convert(model, x,ilens):
  with torch.no_grad():
    scale = 0.1 
    zero_point = 0 
    x_q = torch.quantize_per_tensor(x,scale = scale,zero_point=zero_point,dtype=torch.qint8)
    ilens_q = torch.quantize_per_tensor(ilens.float(),scale = scale,zero_point=zero_point,dtype=torch.qint8)
    nq_si = (x,ilens)
    q_si = (x_q,ilens_q)
    seq_len_dim = Dim("seq_len",min=1,max=1024)
    dynamic_shapes = {
        "x": {0:1,1:seq_len_dim},
        "ilens": {0:1},
    }
    m = export_for_training(model, (x,ilens),dynamic_shapes= dynamic_shapes).module()

    quantizer = XNNPACKQuantizer()
    operator_config = get_symmetric_quantization_config(is_per_channel=False,is_dynamic=True)
    quantizer.set_global(operator_config)
    m = prepare_pt2e(m, quantizer)
    m(*(x,ilens))
    m = convert_pt2e(m)

    exported_program: ExportedProgram  = export(mod=m, args=q_si,dynamic_shapes = dynamic_shapes)
    edge = to_edge_transform_and_lower(
      exported_program,  
      compile_config=EdgeCompileConfig(_check_ir_validity=True),
      partitioner=[XnnpackPartitioner()])
    
    exec_prog = edge.to_executorch()
    print("DONE")

m = MyModule().to('cpu')
m.eval()
x = torch.randn(torch.Size((1,337,80)))
ilens = torch.randn(torch.Size((1,)))
xnnpack_quantize_n_convert(m,x,ilens)

Logs

V0226 15:33:00.389000 2568045 torch/_dynamo/convert_frame.py:1365] skipping: _wrapped_call_impl (reason: in skipfiles, file: python3.10/site-packages/torch/nn/modules/module.py)
V0226 15:33:00.389000 2568045 torch/_dynamo/convert_frame.py:1365] skipping: _call_impl (reason: in skipfiles, file: python3.10/site-packages/torch/nn/modules/module.py)
I0226 15:33:00.392000 2568045 torch/_dynamo/utils.py:1512] [0/0] ChromiumEventLogger initialized with id 56391616-de4c-468d-923a-88bf29db3beb
V0226 15:33:00.396000 2568045 torch/_dynamo/convert_frame.py:941] [0/0] torchdynamo start compiling forward /temp.py:24, stack (elided 4 frames):
V0226 15:33:00.396000 2568045 torch/_dynamo/convert_frame.py:941] [0/0]   File "/temp.py", line 67, in <module>
V0226 15:33:00.396000 2568045 torch/_dynamo/convert_frame.py:941] [0/0]     xnnpack_quantize_n_convert(m,x,ilens)
V0226 15:33:00.396000 2568045 torch/_dynamo/convert_frame.py:941] [0/0]   File "/temp.py", line 45, in xnnpack_quantize_n_convert
V0226 15:33:00.396000 2568045 torch/_dynamo/convert_frame.py:941] [0/0]     m = export_for_training(model, (x,ilens),dynamic_shapes= dynamic_shapes).module()
V0226 15:33:00.396000 2568045 torch/_dynamo/convert_frame.py:941] [0/0]   File "python3.10/site-packages/torch/export/__init__.py", line 168, in export_for_training
V0226 15:33:00.396000 2568045 torch/_dynamo/convert_frame.py:941] [0/0]     return _export_for_training(
V0226 15:33:00.396000 2568045 torch/_dynamo/convert_frame.py:941] [0/0]   File "python3.10/site-packages/torch/export/_trace.py", line 1017, in wrapper
V0226 15:33:00.396000 2568045 torch/_dynamo/convert_frame.py:941] [0/0]     ep = fn(*args, **kwargs)
V0226 15:33:00.396000 2568045 torch/_dynamo/convert_frame.py:941] [0/0]   File "python3.10/site-packages/torch/export/exported_program.py", line 117, in wrapper
V0226 15:33:00.396000 2568045 torch/_dynamo/convert_frame.py:941] [0/0]     return fn(*args, **kwargs)
V0226 15:33:00.396000 2568045 torch/_dynamo/convert_frame.py:941] [0/0]   File "python3.10/site-packages/torch/export/_trace.py", line 1944, in _export_for_training
V0226 15:33:00.396000 2568045 torch/_dynamo/convert_frame.py:941] [0/0]     export_artifact = export_func(  # type: ignore[operator]
V0226 15:33:00.396000 2568045 torch/_dynamo/convert_frame.py:941] [0/0]   File "python3.10/site-packages/torch/export/_trace.py", line 1296, in _strict_export_lower_to_aten_ir
V0226 15:33:00.396000 2568045 torch/_dynamo/convert_frame.py:941] [0/0]     gm_torch_level = _export_to_torch_ir(
V0226 15:33:00.396000 2568045 torch/_dynamo/convert_frame.py:941] [0/0]   File "python3.10/site-packages/torch/export/_trace.py", line 693, in _export_to_torch_ir
V0226 15:33:00.396000 2568045 torch/_dynamo/convert_frame.py:941] [0/0]     gm_torch_level, _ = torch._dynamo.export(
V0226 15:33:00.396000 2568045 torch/_dynamo/convert_frame.py:941] [0/0]   File "python3.10/site-packages/torch/_dynamo/eval_frame.py", line 1579, in inner
V0226 15:33:00.396000 2568045 torch/_dynamo/convert_frame.py:941] [0/0]     result_traced = opt_f(*args, **kwargs)
V0226 15:33:00.396000 2568045 torch/_dynamo/convert_frame.py:941] [0/0]   File "python3.10/site-packages/torch/nn/modules/module.py", line 1749, in _wrapped_call_impl
V0226 15:33:00.396000 2568045 torch/_dynamo/convert_frame.py:941] [0/0]     return self._call_impl(*args, **kwargs)
V0226 15:33:00.396000 2568045 torch/_dynamo/convert_frame.py:941] [0/0]   File "python3.10/site-packages/torch/nn/modules/module.py", line 1760, in _call_impl
V0226 15:33:00.396000 2568045 torch/_dynamo/convert_frame.py:941] [0/0]     return forward_call(*args, **kwargs)
V0226 15:33:00.396000 2568045 torch/_dynamo/convert_frame.py:941] [0/0]   File "python3.10/site-packages/torch/_dynamo/eval_frame.py", line 570, in _fn
V0226 15:33:00.396000 2568045 torch/_dynamo/convert_frame.py:941] [0/0]     return fn(*args, **kwargs)
V0226 15:33:00.396000 2568045 torch/_dynamo/convert_frame.py:941] [0/0]   File "python3.10/site-packages/torch/nn/modules/module.py", line 1749, in _wrapped_call_impl
V0226 15:33:00.396000 2568045 torch/_dynamo/convert_frame.py:941] [0/0]     return self._call_impl(*args, **kwargs)
V0226 15:33:00.396000 2568045 torch/_dynamo/convert_frame.py:941] [0/0] 
I0226 15:33:00.397000 2568045 torch/_dynamo/symbolic_convert.py:2746] [0/0] Step 1: torchdynamo start tracing forward /temp.py:24
I0226 15:33:00.398000 2568045 torch/fx/experimental/symbolic_shapes.py:3288] [0/0] create_env
V0226 15:33:00.408000 2568045 torch/_dynamo/variables/builder.py:2886] [0/0] wrap_to_fake L['x'] (1, 337, 80) StatefulSymbolicContext(dynamic_sizes=[<DimDynamic.DYNAMIC: 0>, <DimDynamic.DYNAMIC: 0>, <DimDynamic.STATIC: 2>], dynamic_strides=[<DimDynamic.INFER_STRIDE: 4>, <DimDynamic.INFER_STRIDE: 4>, <DimDynamic.INFER_STRIDE: 4>], constraint_sizes=[StrictMinMaxConstraint(warn_only=False, vr=VR[1, 1]), StrictMinMaxConstraint(warn_only=False, vr=VR[1, 1024]), None], constraint_strides=[None, None, None], view_base_context=None, tensor_source=LocalSource(local_name='x', is_input=True, is_derefed_cell_contents=False), shape_env_to_source_to_symbol_cache={}) <class 'torch.Tensor'>
V0226 15:33:00.459000 2568045 torch/fx/experimental/symbolic_shapes.py:5967] [0/0] _update_var_to_range s0 = VR[2, 1024] (update)
I0226 15:33:00.459000 2568045 torch/fx/experimental/symbolic_shapes.py:4542] [0/0] create_symbol s0 = 337 for L['x'].size()[1] [2, 1024] (_dynamo/variables/builder.py:2894 in <lambda>), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_CREATE_SYMBOL="s0" or to suppress this message run with TORCHDYNAMO_EXTENDED_ADVICE="0"
V0226 15:33:00.460000 2568045 torch/fx/experimental/symbolic_shapes.py:6781] [0/0] runtime_assert True == True [statically known]
V0226 15:33:00.463000 2568045 torch/_dynamo/output_graph.py:2194] [0/0] create_graph_input L_x_ L['x'] FakeTensor(..., size=(1, s0, 80)) at debug_level 0 before=False
V0226 15:33:00.464000 2568045 torch/_dynamo/variables/builder.py:2886] [0/0] wrap_to_fake L['ilens'] (1,) StatefulSymbolicContext(dynamic_sizes=[<DimDynamic.DYNAMIC: 0>], dynamic_strides=[<DimDynamic.INFER_STRIDE: 4>], constraint_sizes=[StrictMinMaxConstraint(warn_only=False, vr=VR[1, 1])], constraint_strides=[None], view_base_context=None, tensor_source=LocalSource(local_name='ilens', is_input=True, is_derefed_cell_contents=False), shape_env_to_source_to_symbol_cache={}) <class 'torch.Tensor'>
V0226 15:33:00.465000 2568045 torch/_dynamo/output_graph.py:2194] [0/0] create_graph_input L_ilens_ L['ilens'] FakeTensor(..., size=(1,)) at debug_level 0 before=False
V0226 15:33:00.466000 2568045 torch/_dynamo/symbolic_convert.py:958] [0/0] [__trace_source] TRACE starts_line /temp.py:25 in forward (MyModule.forward)
V0226 15:33:00.466000 2568045 torch/_dynamo/symbolic_convert.py:958] [0/0] [__trace_source]             xd = self.fc(x)
V0226 15:33:00.469000 2568045 torch/_dynamo/symbolic_convert.py:981] [0/0] [__trace_bytecode] TRACE LOAD_FAST self []
V0226 15:33:00.469000 2568045 torch/_dynamo/symbolic_convert.py:981] [0/0] [__trace_bytecode] TRACE LOAD_ATTR fc [NNModuleVariable()]
V0226 15:33:00.470000 2568045 torch/_dynamo/symbolic_convert.py:981] [0/0] [__trace_bytecode] TRACE LOAD_FAST x [NNModuleVariable()]
V0226 15:33:00.470000 2568045 torch/_dynamo/symbolic_convert.py:981] [0/0] [__trace_bytecode] TRACE CALL_FUNCTION 1 [NNModuleVariable(), TensorVariable()]
V0226 15:33:00.478000 2568045 torch/fx/experimental/symbolic_shapes.py:6579] [0/0] eval Ne(s0, 1) == True [statically known]
V0226 15:33:00.480000 2568045 torch/fx/experimental/symbolic_shapes.py:6781] [0/0] runtime_assert True == True [statically known]
V0226 15:33:00.481000 2568045 torch/fx/experimental/symbolic_shapes.py:6579] [0/0] eval Eq(s0, 1) == False [statically known]
V0226 15:33:00.485000 2568045 torch/fx/experimental/symbolic_shapes.py:6579] [0/0] eval Ne(Mod(1, s0), 0) == True [statically known]
V0226 15:33:00.497000 2568045 torch/_dynamo/symbolic_convert.py:981] [0/0] [__trace_bytecode] TRACE STORE_FAST xd [TensorVariable()]
V0226 15:33:00.497000 2568045 torch/_dynamo/symbolic_convert.py:958] [0/0] [__trace_source] TRACE starts_line /temp.py:26 in forward (MyModule.forward)
V0226 15:33:00.497000 2568045 torch/_dynamo/symbolic_convert.py:958] [0/0] [__trace_source]             il = ilens
V0226 15:33:00.498000 2568045 torch/_dynamo/symbolic_convert.py:981] [0/0] [__trace_bytecode] TRACE LOAD_FAST ilens []
V0226 15:33:00.498000 2568045 torch/_dynamo/symbolic_convert.py:981] [0/0] [__trace_bytecode] TRACE STORE_FAST il [TensorVariable()]
V0226 15:33:00.498000 2568045 torch/_dynamo/symbolic_convert.py:958] [0/0] [__trace_source] TRACE starts_line /temp.py:27 in forward (MyModule.forward)
V0226 15:33:00.498000 2568045 torch/_dynamo/symbolic_convert.py:958] [0/0] [__trace_source]             return xd,il
V0226 15:33:00.498000 2568045 torch/_dynamo/symbolic_convert.py:981] [0/0] [__trace_bytecode] TRACE LOAD_FAST xd []
V0226 15:33:00.498000 2568045 torch/_dynamo/symbolic_convert.py:981] [0/0] [__trace_bytecode] TRACE LOAD_FAST il [TensorVariable()]
V0226 15:33:00.499000 2568045 torch/_dynamo/symbolic_convert.py:981] [0/0] [__trace_bytecode] TRACE BUILD_TUPLE 2 [TensorVariable(), TensorVariable()]
V0226 15:33:00.499000 2568045 torch/_dynamo/symbolic_convert.py:981] [0/0] [__trace_bytecode] TRACE RETURN_VALUE None [TupleVariable(length=2)]
I0226 15:33:00.499000 2568045 torch/_dynamo/symbolic_convert.py:3067] [0/0] Step 1: torchdynamo done tracing forward (RETURN_VALUE)
V0226 15:33:00.499000 2568045 torch/_dynamo/symbolic_convert.py:3071] [0/0] RETURN_VALUE triggered compile
V0226 15:33:00.499000 2568045 torch/_dynamo/output_graph.py:970] [0/0] COMPILING GRAPH due to GraphCompileReason(reason='return_value', user_stack=[<FrameSummary file /temp.py, line 27 in forward>], graph_break=False)
V0226 15:33:00.502000 2568045 torch/_dynamo/output_graph.py:1355] [0/0] [__graph_code] TRACED GRAPH
V0226 15:33:00.502000 2568045 torch/_dynamo/output_graph.py:1355] [0/0] [__graph_code]  ===== __compiled_fn_1 =====
V0226 15:33:00.502000 2568045 torch/_dynamo/output_graph.py:1355] [0/0] [__graph_code]  python3.10/site-packages/torch/fx/_lazy_graph_module.py class GraphModule(torch.nn.Module):
V0226 15:33:00.502000 2568045 torch/_dynamo/output_graph.py:1355] [0/0] [__graph_code]     def forward(self, L_x_: "f32[1, s0, 80][80*s0, 80, 1]cpu", L_ilens_: "f32[1][1]cpu"):
V0226 15:33:00.502000 2568045 torch/_dynamo/output_graph.py:1355] [0/0] [__graph_code]         l_x_ = L_x_
V0226 15:33:00.502000 2568045 torch/_dynamo/output_graph.py:1355] [0/0] [__graph_code]         l_ilens_ = L_ilens_
V0226 15:33:00.502000 2568045 torch/_dynamo/output_graph.py:1355] [0/0] [__graph_code]         
V0226 15:33:00.502000 2568045 torch/_dynamo/output_graph.py:1355] [0/0] [__graph_code]          # File: /temp.py:25 in forward, code: xd = self.fc(x)
V0226 15:33:00.502000 2568045 torch/_dynamo/output_graph.py:1355] [0/0] [__graph_code]         xd: "f32[1, s0, 10][10*s0, 10, 1]cpu" = self.L__self___fc(l_x_);  l_x_ = None
V0226 15:33:00.502000 2568045 torch/_dynamo/output_graph.py:1355] [0/0] [__graph_code]         return (xd, l_ilens_)
V0226 15:33:00.502000 2568045 torch/_dynamo/output_graph.py:1355] [0/0] [__graph_code]         
V0226 15:33:00.502000 2568045 torch/_dynamo/output_graph.py:1355] [0/0] [__graph_code] 
I0226 15:33:00.503000 2568045 torch/_dynamo/output_graph.py:1462] [0/0] Step 2: calling compiler function dynamo_normalization_capturing_compiler
I0226 15:33:00.503000 2568045 torch/_dynamo/output_graph.py:1467] [0/0] Step 2: done compiler function dynamo_normalization_capturing_compiler
I0226 15:33:00.509000 2568045 torch/fx/experimental/symbolic_shapes.py:4670] [0/0] produce_guards
V0226 15:33:00.510000 2568045 torch/fx/experimental/symbolic_shapes.py:4890] [0/0] track_symint L['x'].size()[0] 1 None
V0226 15:33:00.510000 2568045 torch/fx/experimental/symbolic_shapes.py:4890] [0/0] track_symint L['x'].size()[1] s0 StrictMinMaxConstraint(warn_only=False, vr=VR[1, 1024])
V0226 15:33:00.511000 2568045 torch/fx/experimental/symbolic_shapes.py:4890] [0/0] track_symint L['x'].size()[2] 80 None
V0226 15:33:00.511000 2568045 torch/fx/experimental/symbolic_shapes.py:4890] [0/0] track_symint L['x'].stride()[0] 80*s0 None
V0226 15:33:00.512000 2568045 torch/fx/experimental/symbolic_shapes.py:4890] [0/0] track_symint L['x'].stride()[1] 80 None
V0226 15:33:00.512000 2568045 torch/fx/experimental/symbolic_shapes.py:4890] [0/0] track_symint L['x'].stride()[2] 1 None
V0226 15:33:00.513000 2568045 torch/fx/experimental/symbolic_shapes.py:4890] [0/0] track_symint L['x'].storage_offset() 0 None
V0226 15:33:00.513000 2568045 torch/fx/experimental/symbolic_shapes.py:4890] [0/0] track_symint L['ilens'].size()[0] 1 None
V0226 15:33:00.513000 2568045 torch/fx/experimental/symbolic_shapes.py:4890] [0/0] track_symint L['ilens'].stride()[0] 1 None
V0226 15:33:00.514000 2568045 torch/fx/experimental/symbolic_shapes.py:4890] [0/0] track_symint L['ilens'].storage_offset() 0 None
V0226 15:33:00.525000 2568045 torch/_dynamo/guards.py:1270] [0/0] [__guards] Python shape guard function:
V0226 15:33:00.525000 2568045 torch/_dynamo/guards.py:1270] [0/0] [__guards] def ___make_guard_fn(IsNonOverlappingAndDenseIndicator, cast_symbool_to_symint_guardless, math, torch, ___check_type_id, ___check_obj_id, ___odict_getitem, ___key_to_id, ___dict_version, ___dict_contains, ___tuple_iterator_len, ___normalize_range_iter, ___tuple_iterator_getitem, ___get_torch_function_mode_stack_at, __math_isnan, __numpy_isnan, inf, __load_module, utils_device, device, ___from_numpy, ___as_tensor, inspect):
V0226 15:33:00.525000 2568045 torch/_dynamo/guards.py:1270] [0/0] [__guards]     def guard(L):
V0226 15:33:00.525000 2568045 torch/_dynamo/guards.py:1270] [0/0] [__guards]         _var0 = L['x']
V0226 15:33:00.525000 2568045 torch/_dynamo/guards.py:1270] [0/0] [__guards]         _var1 = _var0.size
V0226 15:33:00.525000 2568045 torch/_dynamo/guards.py:1270] [0/0] [__guards]         _var2 = _var1()
V0226 15:33:00.525000 2568045 torch/_dynamo/guards.py:1270] [0/0] [__guards]         if not (_var2[0] == 1):
V0226 15:33:00.525000 2568045 torch/_dynamo/guards.py:1270] [0/0] [__guards]             return False
V0226 15:33:00.525000 2568045 torch/_dynamo/guards.py:1270] [0/0] [__guards]         if not (_var2[2] == 80):
V0226 15:33:00.525000 2568045 torch/_dynamo/guards.py:1270] [0/0] [__guards]             return False
V0226 15:33:00.525000 2568045 torch/_dynamo/guards.py:1270] [0/0] [__guards]         _var3 = _var0.stride
V0226 15:33:00.525000 2568045 torch/_dynamo/guards.py:1270] [0/0] [__guards]         _var4 = _var3()
V0226 15:33:00.525000 2568045 torch/_dynamo/guards.py:1270] [0/0] [__guards]         _var5 = _var2[1]
V0226 15:33:00.525000 2568045 torch/_dynamo/guards.py:1270] [0/0] [__guards]         if not (_var4[0] == 80 * _var5):
V0226 15:33:00.525000 2568045 torch/_dynamo/guards.py:1270] [0/0] [__guards]             return False
V0226 15:33:00.525000 2568045 torch/_dynamo/guards.py:1270] [0/0] [__guards]         if not (_var4[1] == 80):
V0226 15:33:00.525000 2568045 torch/_dynamo/guards.py:1270] [0/0] [__guards]             return False
V0226 15:33:00.525000 2568045 torch/_dynamo/guards.py:1270] [0/0] [__guards]         if not (_var4[2] == 1):
V0226 15:33:00.525000 2568045 torch/_dynamo/guards.py:1270] [0/0] [__guards]             return False
V0226 15:33:00.525000 2568045 torch/_dynamo/guards.py:1270] [0/0] [__guards]         if not (_var0.storage_offset() == 0):
V0226 15:33:00.525000 2568045 torch/_dynamo/guards.py:1270] [0/0] [__guards]             return False
V0226 15:33:00.525000 2568045 torch/_dynamo/guards.py:1270] [0/0] [__guards]         _var6 = L['ilens']
V0226 15:33:00.525000 2568045 torch/_dynamo/guards.py:1270] [0/0] [__guards]         if not (_var6.size()[0] == 1):
V0226 15:33:00.525000 2568045 torch/_dynamo/guards.py:1270] [0/0] [__guards]             return False
V0226 15:33:00.525000 2568045 torch/_dynamo/guards.py:1270] [0/0] [__guards]         if not (_var6.stride()[0] == 1):
V0226 15:33:00.525000 2568045 torch/_dynamo/guards.py:1270] [0/0] [__guards]             return False
V0226 15:33:00.525000 2568045 torch/_dynamo/guards.py:1270] [0/0] [__guards]         if not (_var6.storage_offset() == 0):
V0226 15:33:00.525000 2568045 torch/_dynamo/guards.py:1270] [0/0] [__guards]             return False
V0226 15:33:00.525000 2568045 torch/_dynamo/guards.py:1270] [0/0] [__guards]         if not (2 <= _var5 and _var5 <= 1024):
V0226 15:33:00.525000 2568045 torch/_dynamo/guards.py:1270] [0/0] [__guards]             return False
V0226 15:33:00.525000 2568045 torch/_dynamo/guards.py:1270] [0/0] [__guards]         return True
V0226 15:33:00.525000 2568045 torch/_dynamo/guards.py:1270] [0/0] [__guards]     return guard
V0226 15:33:00.525000 2568045 torch/_dynamo/guards.py:1270] [0/0] [__guards] 
V0226 15:33:00.525000 2568045 torch/_dynamo/guards.py:2513] [0/0] [__guards] GUARDS:
V0226 15:33:00.526000 2568045 torch/_dynamo/guards.py:2451] [0/0] [__guards] 
V0226 15:33:00.526000 2568045 torch/_dynamo/guards.py:2451] [0/0] [__guards] TREE_GUARD_MANAGER:
V0226 15:33:00.526000 2568045 torch/_dynamo/guards.py:2451] [0/0] [__guards] +- RootGuardManager
V0226 15:33:00.526000 2568045 torch/_dynamo/guards.py:2451] [0/0] [__guards] | +- DEFAULT_DEVICE: utils_device.CURRENT_DEVICE == None                           # _dynamo/output_graph.py:491 in init_ambient_guards
V0226 15:33:00.526000 2568045 torch/_dynamo/guards.py:2451] [0/0] [__guards] | +- GLOBAL_STATE: ___check_global_state()
V0226 15:33:00.526000 2568045 torch/_dynamo/guards.py:2451] [0/0] [__guards] | +- TORCH_FUNCTION_MODE_STACK: ___check_torch_function_mode_stack()
V0226 15:33:00.526000 2568045 torch/_dynamo/guards.py:2451] [0/0] [__guards] | +- GuardManager: source=L['x'], accessed_by=FrameLocalsGuardAccessor(key='x', framelocals_idx=1)
V0226 15:33:00.526000 2568045 torch/_dynamo/guards.py:2451] [0/0] [__guards] | | +- TYPE_MATCH: ___check_type_id(L['x'], 110353158665552)                     # _dynamo/variables/builder.py:1744 in wrap_tensor
V0226 15:33:00.526000 2568045 torch/_dynamo/guards.py:2451] [0/0] [__guards] | | +- DYNAMIC_INDICES: ((L['x']._dynamo_dynamic_indices.issubset(set())) if hasattr(L['x'], '_dynamo_dynamic_indices') else True)  # _dynamo/variables/builder.py:1744 in wrap_tensor
V0226 15:33:00.526000 2568045 torch/_dynamo/guards.py:2451] [0/0] [__guards] | +- GuardManager: source=L['self'], accessed_by=FrameLocalsGuardAccessor(key='self', framelocals_idx=0)
V0226 15:33:00.526000 2568045 torch/_dynamo/guards.py:2451] [0/0] [__guards] | | +- ID_MATCH: ___check_obj_id(L['self'], 125454710324432)                   # _dynamo/output_graph.py:793 in register_attr_or_module
V0226 15:33:00.526000 2568045 torch/_dynamo/guards.py:2451] [0/0] [__guards] | | +- GuardManager: source=L['self'].__dict__, accessed_by=GetGenericDictGuardAccessor
V0226 15:33:00.526000 2568045 torch/_dynamo/guards.py:2451] [0/0] [__guards] | | | +- GuardManager: source=L['self'].training, accessed_by=DictGetItemGuardAccessor('training')
V0226 15:33:00.526000 2568045 torch/_dynamo/guards.py:2451] [0/0] [__guards] | | | | +- ID_MATCH: ___check_obj_id(L['self'].training, 110353092011008)          # _dynamo/output_graph.py:793 in register_attr_or_module
V0226 15:33:00.526000 2568045 torch/_dynamo/guards.py:2451] [0/0] [__guards] | | | +- GuardManager: source=L['self']._modules, accessed_by=DictGetItemGuardAccessor('_modules')
V0226 15:33:00.526000 2568045 torch/_dynamo/guards.py:2451] [0/0] [__guards] | | | | +- GuardManager: source=L['self'].fc, accessed_by=DictGetItemGuardAccessor('fc')
V0226 15:33:00.526000 2568045 torch/_dynamo/guards.py:2451] [0/0] [__guards] | | | | | +- ID_MATCH: ___check_obj_id(L['self'].fc, 125454710324384)                # xd = self.fc(x)  # e2e_pytorch/temp.py:25 in forward
V0226 15:33:00.526000 2568045 torch/_dynamo/guards.py:2451] [0/0] [__guards] | | | | | +- GuardManager: source=L['self'].fc.__dict__, accessed_by=GetGenericDictGuardAccessor
V0226 15:33:00.526000 2568045 torch/_dynamo/guards.py:2451] [0/0] [__guards] | | | | | | +- GuardManager: source=L['self'].fc.training, accessed_by=DictGetItemGuardAccessor('training')
V0226 15:33:00.526000 2568045 torch/_dynamo/guards.py:2451] [0/0] [__guards] | | | | | | | +- ID_MATCH: ___check_obj_id(L['self'].fc.training, 110353092011008)       # xd = self.fc(x)  # e2e_pytorch/temp.py:25 in forward
V0226 15:33:00.526000 2568045 torch/_dynamo/guards.py:2451] [0/0] [__guards] | +- GuardManager: source=L['ilens'], accessed_by=FrameLocalsGuardAccessor(key='ilens', framelocals_idx=2)
V0226 15:33:00.526000 2568045 torch/_dynamo/guards.py:2451] [0/0] [__guards] | | +- TYPE_MATCH: ___check_type_id(L['ilens'], 110353158665552)                 # _dynamo/variables/builder.py:1744 in wrap_tensor
V0226 15:33:00.526000 2568045 torch/_dynamo/guards.py:2451] [0/0] [__guards] | | +- DYNAMIC_INDICES: ((L['ilens']._dynamo_dynamic_indices.issubset(set())) if hasattr(L['ilens'], '_dynamo_dynamic_indices') else True)  # _dynamo/variables/builder.py:1744 in wrap_tensor
V0226 15:33:00.526000 2568045 torch/_dynamo/guards.py:2451] [0/0] [__guards] +- LAMBDA_GUARD: L['x'].size()[0] == 1  # (unknown source L['x'].size()[0], please file a bug)
V0226 15:33:00.526000 2568045 torch/_dynamo/guards.py:2451] [0/0] [__guards] +- LAMBDA_GUARD: L['x'].size()[2] == 80  # (unknown source L['x'].size()[2], please file a bug)
V0226 15:33:00.526000 2568045 torch/_dynamo/guards.py:2451] [0/0] [__guards] +- LAMBDA_GUARD: L['x'].stride()[0] == 80*L['x'].size()[1]  # (unknown source L['x'].stride()[0], please file a bug)
V0226 15:33:00.526000 2568045 torch/_dynamo/guards.py:2451] [0/0] [__guards] +- LAMBDA_GUARD: L['x'].stride()[1] == 80  # (unknown source L['x'].stride()[1], please file a bug)
V0226 15:33:00.526000 2568045 torch/_dynamo/guards.py:2451] [0/0] [__guards] +- LAMBDA_GUARD: L['x'].stride()[2] == 1  # (unknown source L['x'].stride()[2], please file a bug)
V0226 15:33:00.526000 2568045 torch/_dynamo/guards.py:2451] [0/0] [__guards] +- LAMBDA_GUARD: L['x'].storage_offset() == 0  # (unknown source L['x'].storage_offset(), please file a bug)
V0226 15:33:00.526000 2568045 torch/_dynamo/guards.py:2451] [0/0] [__guards] +- LAMBDA_GUARD: L['ilens'].size()[0] == 1  # (unknown source L['ilens'].size()[0], please file a bug)
V0226 15:33:00.526000 2568045 torch/_dynamo/guards.py:2451] [0/0] [__guards] +- LAMBDA_GUARD: L['ilens'].stride()[0] == 1  # (unknown source L['ilens'].stride()[0], please file a bug)
V0226 15:33:00.526000 2568045 torch/_dynamo/guards.py:2451] [0/0] [__guards] +- LAMBDA_GUARD: L['ilens'].storage_offset() == 0  # (unknown source L['ilens'].storage_offset(), please file a bug)
V0226 15:33:00.526000 2568045 torch/_dynamo/guards.py:2451] [0/0] [__guards] +- LAMBDA_GUARD: 2 <= L['x'].size()[1] <= 1024  # (user code shown is first use of this value--the guard itself is not due user code but due to 0/1 specialization in the framework; to avoid specialization try torch._dynamo.mark_unbacked(tensor, dim)) and (_dynamo/variables/builder.py:2894 in <lambda>)
V0226 15:33:00.526000 2568045 torch/_dynamo/guards.py:2451] [0/0] [__guards] 
I0226 15:33:00.527000 2568045 torch/_dynamo/pgo.py:647] [0/0] put_code_state: no cache key, skipping
I0226 15:33:00.527000 2568045 torch/_dynamo/convert_frame.py:1059] [0/0] run_gc_after_compile: running gc
V0226 15:33:00.530000 2568045 torch/_dynamo/convert_frame.py:1365] skipping: _fn (reason: in skipfiles, file: python3.10/site-packages/torch/_dynamo/eval_frame.py)
I0226 15:33:00.546000 2568045 torch/_dynamo/eval_frame.py:1609] Summary of dimension constraints:
I0226 15:33:00.546000 2568045 torch/_dynamo/eval_frame.py:1665] Dynamo captured graph:
I0226 15:33:00.546000 2568045 torch/_dynamo/eval_frame.py:1665] 
I0226 15:33:00.546000 2568045 torch/_dynamo/eval_frame.py:1665] class GraphModule(torch.nn.Module):
I0226 15:33:00.546000 2568045 torch/_dynamo/eval_frame.py:1665]     def forward(self, L_x_: "f32[1, s0, 80]", L_ilens_: "f32[1]"):
I0226 15:33:00.546000 2568045 torch/_dynamo/eval_frame.py:1665]         l_x_ = L_x_
I0226 15:33:00.546000 2568045 torch/_dynamo/eval_frame.py:1665]         l_ilens_ = L_ilens_
I0226 15:33:00.546000 2568045 torch/_dynamo/eval_frame.py:1665]         
I0226 15:33:00.546000 2568045 torch/_dynamo/eval_frame.py:1665]          # File: /temp.py:25 in forward, code: xd = self.fc(x)
I0226 15:33:00.546000 2568045 torch/_dynamo/eval_frame.py:1665]         xd: "f32[1, s0, 10]" = self.L__self___fc(l_x_);  l_x_ = None
I0226 15:33:00.546000 2568045 torch/_dynamo/eval_frame.py:1665]         return (xd, l_ilens_)
I0226 15:33:00.546000 2568045 torch/_dynamo/eval_frame.py:1665]         
V0226 15:33:00.591000 2568045 torch/_dynamo/convert_frame.py:1365] skipping: call_wrapped (reason: in skipfiles, file: python3.10/site-packages/torch/fx/graph_module.py)
V0226 15:33:00.591000 2568045 torch/_dynamo/convert_frame.py:1365] skipping: __call__ (reason: in skipfiles, file: python3.10/site-packages/torch/fx/graph_module.py)
V0226 15:33:00.592000 2568045 torch/_dynamo/convert_frame.py:941] [1/0] torchdynamo start compiling forward <eval_with_key>.29:4, stack (elided 4 frames):
V0226 15:33:00.592000 2568045 torch/_dynamo/convert_frame.py:941] [1/0]   File "/temp.py", line 67, in <module>
V0226 15:33:00.592000 2568045 torch/_dynamo/convert_frame.py:941] [1/0]     xnnpack_quantize_n_convert(m,x,ilens)
V0226 15:33:00.592000 2568045 torch/_dynamo/convert_frame.py:941] [1/0]   File "/temp.py", line 54, in xnnpack_quantize_n_convert
V0226 15:33:00.592000 2568045 torch/_dynamo/convert_frame.py:941] [1/0]     exported_program: ExportedProgram  = export(mod=m, args=q_si,dynamic_shapes = dynamic_shapes)
V0226 15:33:00.592000 2568045 torch/_dynamo/convert_frame.py:941] [1/0]   File "python3.10/site-packages/torch/export/__init__.py", line 368, in export
V0226 15:33:00.592000 2568045 torch/_dynamo/convert_frame.py:941] [1/0]     return _export(
V0226 15:33:00.592000 2568045 torch/_dynamo/convert_frame.py:941] [1/0]   File "python3.10/site-packages/torch/export/_trace.py", line 1017, in wrapper
V0226 15:33:00.592000 2568045 torch/_dynamo/convert_frame.py:941] [1/0]     ep = fn(*args, **kwargs)
V0226 15:33:00.592000 2568045 torch/_dynamo/convert_frame.py:941] [1/0]   File "python3.10/site-packages/torch/export/exported_program.py", line 117, in wrapper
V0226 15:33:00.592000 2568045 torch/_dynamo/convert_frame.py:941] [1/0]     return fn(*args, **kwargs)
V0226 15:33:00.592000 2568045 torch/_dynamo/convert_frame.py:941] [1/0]   File "python3.10/site-packages/torch/export/_trace.py", line 2079, in _export
V0226 15:33:00.592000 2568045 torch/_dynamo/convert_frame.py:941] [1/0]     return _export_for_training(
V0226 15:33:00.592000 2568045 torch/_dynamo/convert_frame.py:941] [1/0]   File "python3.10/site-packages/torch/export/_trace.py", line 1017, in wrapper
V0226 15:33:00.592000 2568045 torch/_dynamo/convert_frame.py:941] [1/0]     ep = fn(*args, **kwargs)
V0226 15:33:00.592000 2568045 torch/_dynamo/convert_frame.py:941] [1/0]   File "python3.10/site-packages/torch/export/exported_program.py", line 117, in wrapper
V0226 15:33:00.592000 2568045 torch/_dynamo/convert_frame.py:941] [1/0]     return fn(*args, **kwargs)
V0226 15:33:00.592000 2568045 torch/_dynamo/convert_frame.py:941] [1/0]   File "python3.10/site-packages/torch/export/_trace.py", line 1944, in _export_for_training
V0226 15:33:00.592000 2568045 torch/_dynamo/convert_frame.py:941] [1/0]     export_artifact = export_func(  # type: ignore[operator]
V0226 15:33:00.592000 2568045 torch/_dynamo/convert_frame.py:941] [1/0]   File "python3.10/site-packages/torch/export/_trace.py", line 1296, in _strict_export_lower_to_aten_ir
V0226 15:33:00.592000 2568045 torch/_dynamo/convert_frame.py:941] [1/0]     gm_torch_level = _export_to_torch_ir(
V0226 15:33:00.592000 2568045 torch/_dynamo/convert_frame.py:941] [1/0]   File "python3.10/site-packages/torch/export/_trace.py", line 693, in _export_to_torch_ir
V0226 15:33:00.592000 2568045 torch/_dynamo/convert_frame.py:941] [1/0]     gm_torch_level, _ = torch._dynamo.export(
V0226 15:33:00.592000 2568045 torch/_dynamo/convert_frame.py:941] [1/0]   File "python3.10/site-packages/torch/_dynamo/eval_frame.py", line 1579, in inner
V0226 15:33:00.592000 2568045 torch/_dynamo/convert_frame.py:941] [1/0]     result_traced = opt_f(*args, **kwargs)
V0226 15:33:00.592000 2568045 torch/_dynamo/convert_frame.py:941] [1/0]   File "python3.10/site-packages/torch/nn/modules/module.py", line 1749, in _wrapped_call_impl
V0226 15:33:00.592000 2568045 torch/_dynamo/convert_frame.py:941] [1/0]     return self._call_impl(*args, **kwargs)
V0226 15:33:00.592000 2568045 torch/_dynamo/convert_frame.py:941] [1/0]   File "python3.10/site-packages/torch/nn/modules/module.py", line 1760, in _call_impl
V0226 15:33:00.592000 2568045 torch/_dynamo/convert_frame.py:941] [1/0]     return forward_call(*args, **kwargs)
V0226 15:33:00.592000 2568045 torch/_dynamo/convert_frame.py:941] [1/0]   File "python3.10/site-packages/torch/_dynamo/eval_frame.py", line 570, in _fn
V0226 15:33:00.592000 2568045 torch/_dynamo/convert_frame.py:941] [1/0]     return fn(*args, **kwargs)
V0226 15:33:00.592000 2568045 torch/_dynamo/convert_frame.py:941] [1/0]   File "python3.10/site-packages/torch/fx/graph_module.py", line 824, in call_wrapped
V0226 15:33:00.592000 2568045 torch/_dynamo/convert_frame.py:941] [1/0]     return self._wrapped_call(self, *args, **kwargs)
V0226 15:33:00.592000 2568045 torch/_dynamo/convert_frame.py:941] [1/0]   File "python3.10/site-packages/torch/fx/graph_module.py", line 387, in __call__
V0226 15:33:00.592000 2568045 torch/_dynamo/convert_frame.py:941] [1/0]     return super(self.cls, obj).__call__(*args, **kwargs)  # type: ignore[misc]
V0226 15:33:00.592000 2568045 torch/_dynamo/convert_frame.py:941] [1/0]   File "python3.10/site-packages/torch/nn/modules/module.py", line 1749, in _wrapped_call_impl
V0226 15:33:00.592000 2568045 torch/_dynamo/convert_frame.py:941] [1/0]     return self._call_impl(*args, **kwargs)
V0226 15:33:00.592000 2568045 torch/_dynamo/convert_frame.py:941] [1/0] 
I0226 15:33:00.594000 2568045 torch/_dynamo/symbolic_convert.py:2746] [1/0] Step 1: torchdynamo start tracing forward <eval_with_key>.29:4
I0226 15:33:00.595000 2568045 torch/fx/experimental/symbolic_shapes.py:3288] [1/0] create_env
V0226 15:33:00.596000 2568045 torch/_dynamo/variables/builder.py:2886] [1/0] wrap_to_fake L['x'] (1, 337, 80) StatefulSymbolicContext(dynamic_sizes=[<DimDynamic.DYNAMIC: 0>, <DimDynamic.DYNAMIC: 0>, <DimDynamic.STATIC: 2>], dynamic_strides=[<DimDynamic.INFER_STRIDE: 4>, <DimDynamic.INFER_STRIDE: 4>, <DimDynamic.INFER_STRIDE: 4>], constraint_sizes=[StrictMinMaxConstraint(warn_only=False, vr=VR[1, 1]), StrictMinMaxConstraint(warn_only=False, vr=VR[1, 1024]), None], constraint_strides=[None, None, None], view_base_context=None, tensor_source=LocalSource(local_name='x', is_input=True, is_derefed_cell_contents=False), shape_env_to_source_to_symbol_cache={}) <class 'torch.Tensor'>
W0226 15:33:00.597000 2568045 torch/_dynamo/utils.py:2512] [1/0] Unsupported: quantized nyi in meta tensors with fake tensor propagation.
I0226 15:33:00.597000 2568045 torch/_dynamo/convert_frame.py:1059] [1/0] run_gc_after_compile: running gc
Traceback (most recent call last):
  File "python3.10/site-packages/torch/_dynamo/utils.py", line 2507, in wrap_fake_exception
    return fn()
  File "python3.10/site-packages/torch/_dynamo/variables/builder.py", line 2894, in <lambda>
    lambda: tx.fake_mode.from_tensor(
  File "python3.10/site-packages/torch/_subclasses/fake_tensor.py", line 2604, in from_tensor
    return self.fake_tensor_converter.from_real_tensor(
  File "python3.10/site-packages/torch/_subclasses/fake_tensor.py", line 350, in from_real_tensor
    raise UnsupportedFakeTensorException("quantized nyi in meta tensors")
torch._subclasses.fake_tensor.UnsupportedFakeTensorException: quantized nyi in meta tensors

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
  File "/temp.py", line 67, in <module>
    xnnpack_quantize_n_convert(m,x,ilens)
  File "/temp.py", line 54, in xnnpack_quantize_n_convert
    exported_program: ExportedProgram  = export(mod=m, args=q_si,dynamic_shapes = dynamic_shapes)
  File "python3.10/site-packages/torch/export/__init__.py", line 368, in export
    return _export(
  File "python3.10/site-packages/torch/export/_trace.py", line 1044, in wrapper
    raise e
  File "python3.10/site-packages/torch/export/_trace.py", line 1017, in wrapper
    ep = fn(*args, **kwargs)
  File "python3.10/site-packages/torch/export/exported_program.py", line 117, in wrapper
    return fn(*args, **kwargs)
  File "python3.10/site-packages/torch/export/_trace.py", line 2079, in _export
    return _export_for_training(
  File "python3.10/site-packages/torch/export/_trace.py", line 1044, in wrapper
    raise e
  File "python3.10/site-packages/torch/export/_trace.py", line 1017, in wrapper
    ep = fn(*args, **kwargs)
  File "python3.10/site-packages/torch/export/exported_program.py", line 117, in wrapper
    return fn(*args, **kwargs)
  File "python3.10/site-packages/torch/export/_trace.py", line 1944, in _export_for_training
    export_artifact = export_func(  # type: ignore[operator]
  File "python3.10/site-packages/torch/export/_trace.py", line 1296, in _strict_export_lower_to_aten_ir
    gm_torch_level = _export_to_torch_ir(
  File "python3.10/site-packages/torch/export/_trace.py", line 693, in _export_to_torch_ir
    gm_torch_level, _ = torch._dynamo.export(
  File "python3.10/site-packages/torch/_dynamo/eval_frame.py", line 1579, in inner
    result_traced = opt_f(*args, **kwargs)
  File "python3.10/site-packages/torch/nn/modules/module.py", line 1749, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
  File "python3.10/site-packages/torch/nn/modules/module.py", line 1760, in _call_impl
    return forward_call(*args, **kwargs)
  File "python3.10/site-packages/torch/_dynamo/eval_frame.py", line 570, in _fn
    return fn(*args, **kwargs)
  File "python3.10/site-packages/torch/fx/graph_module.py", line 824, in call_wrapped
    return self._wrapped_call(self, *args, **kwargs)
  File "python3.10/site-packages/torch/fx/graph_module.py", line 400, in __call__
    raise e
  File "python3.10/site-packages/torch/fx/graph_module.py", line 387, in __call__
    return super(self.cls, obj).__call__(*args, **kwargs)  # type: ignore[misc]
  File "python3.10/site-packages/torch/nn/modules/module.py", line 1749, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
  File "python3.10/site-packages/torch/nn/modules/module.py", line 1760, in _call_impl
    return forward_call(*args, **kwargs)
  File "python3.10/site-packages/torch/_dynamo/convert_frame.py", line 1400, in __call__
    return self._torchdynamo_orig_callable(
  File "python3.10/site-packages/torch/_dynamo/convert_frame.py", line 565, in __call__
    return _compile(
  File "python3.10/site-packages/torch/_dynamo/convert_frame.py", line 997, in _compile
    guarded_code = compile_inner(code, one_graph, hooks, transform)
  File "python3.10/site-packages/torch/_utils_internal.py", line 95, in wrapper_function
    return function(*args, **kwargs)
  File "python3.10/site-packages/torch/_dynamo/convert_frame.py", line 726, in compile_inner
    return _compile_inner(code, one_graph, hooks, transform)
  File "python3.10/site-packages/torch/_dynamo/convert_frame.py", line 760, in _compile_inner
    out_code = transform_code_object(code, transform)
  File "python3.10/site-packages/torch/_dynamo/bytecode_transformation.py", line 1404, in transform_code_object
    transformations(instructions, code_options)
  File "python3.10/site-packages/torch/_dynamo/convert_frame.py", line 236, in _fn
    return fn(*args, **kwargs)
  File "python3.10/site-packages/torch/_dynamo/convert_frame.py", line 660, in transform
    tracer = InstructionTranslator(
  File "python3.10/site-packages/torch/_dynamo/symbolic_convert.py", line 2868, in __init__
    self.symbolic_locals = variables.LazyVariableTracker.realize_all(
  File "python3.10/site-packages/torch/_dynamo/variables/lazy.py", line 139, in realize_all
    result = {k: cls.realize_all(v, cache) for k, v in list(value.items())}
  File "python3.10/site-packages/torch/_dynamo/variables/lazy.py", line 139, in <dictcomp>
    result = {k: cls.realize_all(v, cache) for k, v in list(value.items())}
  File "python3.10/site-packages/torch/_dynamo/variables/lazy.py", line 125, in realize_all
    result = cls.realize_all(value.realize(), cache)
  File "python3.10/site-packages/torch/_dynamo/variables/lazy.py", line 67, in realize
    self._cache.realize()
  File "python3.10/site-packages/torch/_dynamo/variables/lazy.py", line 33, in realize
    self.vt = VariableTracker.build(tx, self.value, source)
  File "python3.10/site-packages/torch/_dynamo/variables/base.py", line 456, in build
    return builder.VariableBuilder(tx, source)(value)
  File "python3.10/site-packages/torch/_dynamo/variables/builder.py", line 384, in __call__
    vt = self._wrap(value)
  File "python3.10/site-packages/torch/_dynamo/variables/builder.py", line 548, in _wrap
    return type_dispatch(self, value)
  File "python3.10/site-packages/torch/_dynamo/variables/builder.py", line 1708, in wrap_tensor
    example_value = wrap_to_fake_tensor_and_record(
  File "python3.10/site-packages/torch/_dynamo/variables/builder.py", line 2893, in wrap_to_fake_tensor_and_record
    fake_e = wrap_fake_exception(
  File "python3.10/site-packages/torch/_dynamo/utils.py", line 2513, in wrap_fake_exception
    unimplemented(msg, from_exc=e)
  File "python3.10/site-packages/torch/_dynamo/exc.py", line 379, in unimplemented
    raise Unsupported(msg, case_name=case_name) from from_exc
torch._dynamo.exc.Unsupported: Unsupported: quantized nyi in meta tensors with fake tensor propagation.

I0226 15:33:04.084000 2568109 torch/_dynamo/eval_frame.py:392] TorchDynamo attempted to trace the following frames: [
I0226 15:33:04.084000 2568109 torch/_dynamo/eval_frame.py:392] 
I0226 15:33:04.084000 2568109 torch/_dynamo/eval_frame.py:392] ]
I0226 15:33:04.094000 2568109 torch/_dynamo/utils.py:746] TorchDynamo compilation metrics:
I0226 15:33:04.094000 2568109 torch/_dynamo/utils.py:746] Function    Runtimes (s)
I0226 15:33:04.094000 2568109 torch/_dynamo/utils.py:746] ----------  --------------
V0226 15:33:04.094000 2568109 torch/fx/experimental/symbolic_shapes.py:164] lru_cache_stats constrain_symbol_range: CacheInfo(hits=0, misses=0, maxsize=None, currsize=0)
V0226 15:33:04.095000 2568109 torch/fx/experimental/symbolic_shapes.py:164] lru_cache_stats defer_runtime_assert: CacheInfo(hits=0, misses=0, maxsize=256, currsize=0)
V0226 15:33:04.095000 2568109 torch/fx/experimental/symbolic_shapes.py:164] lru_cache_stats evaluate_expr: CacheInfo(hits=0, misses=0, maxsize=256, currsize=0)
V0226 15:33:04.095000 2568109 torch/fx/experimental/symbolic_shapes.py:164] lru_cache_stats _simplify_floor_div: CacheInfo(hits=0, misses=0, maxsize=None, currsize=0)
V0226 15:33:04.096000 2568109 torch/fx/experimental/symbolic_shapes.py:164] lru_cache_stats _maybe_guard_rel: CacheInfo(hits=0, misses=0, maxsize=256, currsize=0)
V0226 15:33:04.096000 2568109 torch/fx/experimental/symbolic_shapes.py:164] lru_cache_stats _find: CacheInfo(hits=0, misses=0, maxsize=None, currsize=0)
V0226 15:33:04.097000 2568109 torch/fx/experimental/symbolic_shapes.py:164] lru_cache_stats has_hint: CacheInfo(hits=0, misses=0, maxsize=256, currsize=0)
V0226 15:33:04.097000 2568109 torch/fx/experimental/symbolic_shapes.py:164] lru_cache_stats size_hint: CacheInfo(hits=0, misses=0, maxsize=256, currsize=0)
V0226 15:33:04.097000 2568109 torch/fx/experimental/symbolic_shapes.py:164] lru_cache_stats simplify: CacheInfo(hits=0, misses=0, maxsize=None, currsize=0)
V0226 15:33:04.098000 2568109 torch/fx/experimental/symbolic_shapes.py:164] lru_cache_stats _update_divisible: CacheInfo(hits=0, misses=0, maxsize=None, currsize=0)
V0226 15:33:04.098000 2568109 torch/fx/experimental/symbolic_shapes.py:164] lru_cache_stats replace: CacheInfo(hits=0, misses=0, maxsize=None, currsize=0)
V0226 15:33:04.098000 2568109 torch/fx/experimental/symbolic_shapes.py:164] lru_cache_stats _maybe_evaluate_static: CacheInfo(hits=0, misses=0, maxsize=None, currsize=0)
V0226 15:33:04.099000 2568109 torch/fx/experimental/symbolic_shapes.py:164] lru_cache_stats get_implications: CacheInfo(hits=0, misses=0, maxsize=None, currsize=0)
V0226 15:33:04.099000 2568109 torch/fx/experimental/symbolic_shapes.py:164] lru_cache_stats get_axioms: CacheInfo(hits=0, misses=0, maxsize=None, currsize=0)
V0226 15:33:04.099000 2568109 torch/fx/experimental/symbolic_shapes.py:164] lru_cache_stats _maybe_evaluate_static_worker: CacheInfo(hits=0, misses=0, maxsize=None, currsize=0)
V0226 15:33:04.100000 2568109 torch/fx/experimental/symbolic_shapes.py:164] lru_cache_stats safe_expand: CacheInfo(hits=0, misses=0, maxsize=256, currsize=0)
V0226 15:33:04.100000 2568109 torch/fx/experimental/symbolic_shapes.py:164] lru_cache_stats uninteresting_files: CacheInfo(hits=0, misses=0, maxsize=None, currsize=0)
I0226 15:33:04.705000 2568045 torch/_dynamo/eval_frame.py:392] TorchDynamo attempted to trace the following frames: [
I0226 15:33:04.705000 2568045 torch/_dynamo/eval_frame.py:392]   * forward /temp.py:24
I0226 15:33:04.705000 2568045 torch/_dynamo/eval_frame.py:392]   * forward <eval_with_key>.29:4
I0226 15:33:04.705000 2568045 torch/_dynamo/eval_frame.py:392] ]
I0226 15:33:04.884000 2568045 torch/_dynamo/utils.py:746] TorchDynamo compilation metrics:
I0226 15:33:04.884000 2568045 torch/_dynamo/utils.py:746] Function                          Runtimes (s)
I0226 15:33:04.884000 2568045 torch/_dynamo/utils.py:746] ------------------------------  --------------
I0226 15:33:04.884000 2568045 torch/_dynamo/utils.py:746] _compile.compile_inner                  0.1343
I0226 15:33:04.884000 2568045 torch/_dynamo/utils.py:746] OutputGraph.call_user_compiler          0.0006
I0226 15:33:04.884000 2568045 torch/_dynamo/utils.py:746] gc                                      0.0012
V0226 15:33:04.884000 2568045 torch/fx/experimental/symbolic_shapes.py:164] lru_cache_stats constrain_symbol_range: CacheInfo(hits=0, misses=0, maxsize=None, currsize=0)
V0226 15:33:04.884000 2568045 torch/fx/experimental/symbolic_shapes.py:164] lru_cache_stats defer_runtime_assert: CacheInfo(hits=20, misses=2, maxsize=256, currsize=2)
V0226 15:33:04.884000 2568045 torch/fx/experimental/symbolic_shapes.py:164] lru_cache_stats evaluate_expr: CacheInfo(hits=98, misses=8, maxsize=256, currsize=8)
V0226 15:33:04.884000 2568045 torch/fx/experimental/symbolic_shapes.py:164] lru_cache_stats _simplify_floor_div: CacheInfo(hits=0, misses=0, maxsize=None, currsize=0)
V0226 15:33:04.885000 2568045 torch/fx/experimental/symbolic_shapes.py:164] lru_cache_stats _maybe_guard_rel: CacheInfo(hits=0, misses=0, maxsize=256, currsize=0)
V0226 15:33:04.885000 2568045 torch/fx/experimental/symbolic_shapes.py:164] lru_cache_stats _find: CacheInfo(hits=9, misses=1, maxsize=None, currsize=1)
V0226 15:33:04.885000 2568045 torch/fx/experimental/symbolic_shapes.py:164] lru_cache_stats has_hint: CacheInfo(hits=0, misses=0, maxsize=256, currsize=0)
V0226 15:33:04.885000 2568045 torch/fx/experimental/symbolic_shapes.py:164] lru_cache_stats size_hint: CacheInfo(hits=0, misses=0, maxsize=256, currsize=0)
V0226 15:33:04.885000 2568045 torch/fx/experimental/symbolic_shapes.py:164] lru_cache_stats simplify: CacheInfo(hits=0, misses=4, maxsize=None, currsize=4)
V0226 15:33:04.885000 2568045 torch/fx/experimental/symbolic_shapes.py:164] lru_cache_stats _update_divisible: CacheInfo(hits=0, misses=0, maxsize=None, currsize=0)
V0226 15:33:04.886000 2568045 torch/fx/experimental/symbolic_shapes.py:164] lru_cache_stats replace: CacheInfo(hits=1266, misses=19, maxsize=None, currsize=19)
V0226 15:33:04.886000 2568045 torch/fx/experimental/symbolic_shapes.py:164] lru_cache_stats _maybe_evaluate_static: CacheInfo(hits=1, misses=4, maxsize=None, currsize=4)
V0226 15:33:04.886000 2568045 torch/fx/experimental/symbolic_shapes.py:164] lru_cache_stats get_implications: CacheInfo(hits=0, misses=0, maxsize=None, currsize=0)
V0226 15:33:04.886000 2568045 torch/fx/experimental/symbolic_shapes.py:164] lru_cache_stats get_axioms: CacheInfo(hits=0, misses=0, maxsize=None, currsize=0)
V0226 15:33:04.886000 2568045 torch/fx/experimental/symbolic_shapes.py:164] lru_cache_stats _maybe_evaluate_static_worker: CacheInfo(hits=0, misses=3, maxsize=None, currsize=3)
V0226 15:33:04.886000 2568045 torch/fx/experimental/symbolic_shapes.py:164] lru_cache_stats safe_expand: CacheInfo(hits=2, misses=5, maxsize=256, currsize=5)
V0226 15:33:04.887000 2568045 torch/fx/experimental/symbolic_shapes.py:164] lru_cache_stats uninteresting_files: CacheInfo(hits=62, misses=1, maxsize=None, currsize=1)

Versions

PyTorch version: 2.7.0.dev20250131+cpu
Is debug build: False
CUDA used to build PyTorch: Could not collect
ROCM used to build PyTorch: N/A

OS: Ubuntu 22.04.5 LTS (x86_64)
GCC version: (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0
Clang version: 14.0.0-1ubuntu1.1
CMake version: version 3.31.4
Libc version: glibc-2.35

Python version: 3.10.12 (main, Sep 11 2024, 15:47:36) [GCC 11.4.0] (64-bit runtime)
Python platform: Linux-6.8.0-45-generic-x86_64-with-glibc2.35
Is CUDA available: False
CUDA runtime version: 11.5.119
CUDA_MODULE_LOADING set to: N/A
GPU models and configuration: Could not collect
Nvidia driver version: Could not collect
cuDNN version: Could not collect
HIP runtime version: N/A
MIOpen runtime version: N/A
Is XNNPACK available: True


Versions of relevant libraries:
[pip3] ai-edge-torch==0.3.0
[pip3] executorch==0.6.0a0+b5344c1
[pip3] numpy==2.1.3
[pip3] nvidia-cublas-cu12==12.4.5.8
[pip3] nvidia-cuda-cupti-cu12==12.4.127
[pip3] nvidia-cuda-nvrtc-cu12==12.4.127
[pip3] nvidia-cuda-runtime-cu12==12.4.127
[pip3] nvidia-cudnn-cu12==9.1.0.70
[pip3] nvidia-cufft-cu12==11.2.1.3
[pip3] nvidia-curand-cu12==10.3.5.147
[pip3] nvidia-cusolver-cu12==11.6.1.9
[pip3] nvidia-cusparse-cu12==12.3.1.170
[pip3] nvidia-cusparselt-cu12==0.6.2
[pip3] nvidia-nccl-cu12==2.21.5
[pip3] nvidia-nvjitlink-cu12==12.4.127
[pip3] nvidia-nvtx-cu12==12.4.127
[pip3] optree==0.14.0
[pip3] torch==2.7.0.dev20250131+cpu
[pip3] torch_xla2==0.0.1.dev202412041639
[pip3] torchao==0.8.0+git11333ba2
[pip3] torchaudio==2.6.0.dev20250131+cpu
[pip3] torchsr==1.0.4
[pip3] torchvision==0.22.0.dev20250131+cpu
[pip3] triton==3.2.0

cc @digantdesai @mcr229 @cbilgin @mergennachin @byjlw

[mergennachin]

cc @digantdesai @mcr229 @tarun292 @GregoryComer

[tarun292]

@gpchowdari i'm assuming you're trying to pass in quantized inputs into the model. If so the way you're trying to do it by configuring it as qint8 inputs is not very well supported. The ideal way to do this in ExecuTorch is to use this pass: https://github.com/pytorch/executorch/blob/main/exir/passes/quantize_io_pass.py

You can take a look at this code to see how it's invoked: https://github.com/pytorch/executorch/blob/main/exir/tests/test_quantize_io_pass.py

[gpchowdari]

@gpchowdari i'm assuming you're trying to pass in quantized inputs into the model. If so the way you're trying to do it by configuring it as qint8 inputs is not very well supported. The ideal way to do this in ExecuTorch is to use this pass: https://github.com/pytorch/executorch/blob/main/exir/passes/quantize_io_pass.py

You can take a look at this code to see how it's invoked: https://github.com/pytorch/executorch/blob/main/exir/tests/test_quantize_io_pass.py

@tarun292 Thank you for the reply.

Context:

I am trying to bring one of our multimodal LLM to on-device. I am able to export the model without quantization.
When i try to quantize the model, i am facing the issues.

1. I have followed the steps @https://pytorch.org/executorch/stable/tutorial-xnnpack-delegate-lowering.html to quantize the model. with to_edge_transform_and_lower I have encountered the issue RuntimeError: Weight and Input should both be quantized
2. I have tried with passing the quantized inputs for which have received the error quantized nyi in meta tensors with fake tensor propagation

I will go through the provided references. Thanks.

[tarun292]

@gpchowdari oh if that's the error you are seeing then i don't think this is the right fix. Can you share the whole stack trace of the error you're seeing and the export code that you used (without input quantization etc.).

[gpchowdari]

@tarun292

This is the function used to export and lower with XNNPack backend

def xnnpack_quantize_n_convert(model, x,ilens):
  with torch.no_grad():
    nq_si = (x,ilens)
    seq_len_dim = Dim("seq_len",min=1,max=1024)
    dynamic_shapes = {
        "x": {0:1,1:seq_len_dim},
        "ilens": {0:1},
    }
    m = export_for_training(model, nq_si,dynamic_shapes= dynamic_shapes).module()

    quantizer = XNNPACKQuantizer()
    operator_config = get_symmetric_quantization_config(is_per_channel=False,is_dynamic=True)
    quantizer.set_global(operator_config)
    m = prepare_pt2e(m, quantizer)
    m(*nq_si)
    m = convert_pt2e(m)
    compile_config = get_xnnpack_edge_compile_config()
    exported_program: ExportedProgram  = export(mod=m, args=nq_si,dynamic_shapes = dynamic_shapes)
    edge = to_edge_transform_and_lower(
      exported_program,  
      compile_config=compile_config,
      partitioner=[XnnpackPartitioner()])
    
    exec_prog = edge.to_executorch()

    with open("qs8_xnnpack.pte", "wb") as file:
         exec_prog.write_to_file(file)

stack

V0227 19:46:19.352000 3411345 torch/fx/experimental/symbolic_shapes.py:6579] eval Ne((s1 - 1)*(2*s1 - 3), 1) == True [statically known]
Traceback (most recent call last):
  edge = to_edge_transform_and_lower(
  File "/python3.10/site-packages/executorch/exir/program/_program.py", line 101, in wrapper
    return func(self, *args, **kwargs)
  File "/python3.10/site-packages/executorch/exir/program/_program.py", line 1107, in to_edge_transform_and_lower
    edge_manager = edge_manager.to_backend({name: curr_partitioner})
  File "/python3.10/site-packages/executorch/exir/program/_program.py", line 101, in wrapper
    return func(self, *args, **kwargs)
  File "/python3.10/site-packages/executorch/exir/program/_program.py", line 1362, in to_backend
    new_edge_programs[name] = to_backend(program, partitioner[name])
  File "/usr/lib/python3.10/functools.py", line 889, in wrapper
    return dispatch(args[0].__class__)(*args, **kw)
  File "/python3.10/site-packages/executorch/exir/backend/backend_api.py", line 396, in _
    tagged_graph_module = _partition_and_lower(
  File "/python3.10/site-packages/executorch/exir/backend/backend_api.py", line 319, in _partition_and_lower
    partitioned_module = _partition_and_lower_one_graph_module(
  File "/python3.10/site-packages/executorch/exir/backend/backend_api.py", line 249, in _partition_and_lower_one_graph_module
    lowered_submodule = to_backend(
  File "/usr/lib/python3.10/functools.py", line 889, in wrapper
    return dispatch(args[0].__class__)(*args, **kw)
  File "/python3.10/site-packages/executorch/exir/backend/backend_api.py", line 113, in _
    preprocess_result: PreprocessResult = cls.preprocess(
  File "/python3.10/site-packages/executorch/backends/xnnpack/xnnpack_preprocess.py", line 171, in preprocess
    node_visitors[node.target.__name__].define_node(
  File "/python3.10/site-packages/executorch/backends/xnnpack/operators/op_linear.py", line 69, in define_node
    bias_quant_params = QuantParams.from_bias(
  File "/python3.10/site-packages/executorch/backends/xnnpack/operators/quant_params.py", line 306, in from_bias
    check_or_raise(
  File "/python3.10/site-packages/executorch/backends/xnnpack/utils/utils.py", line 62, in check_or_raise
    raise RuntimeError(err)
RuntimeError: Weight and Input should both be quantized

[tarun292]

Seems reasonable to me overall tagging @mcr229 @digantdesai who might be able to help.

[mcr229]

@gpchowdari if you're using dynamic quantization you might have to use the following pass:

executorch/examples/models/phi-3-mini/export_phi-3-mini.py

Line 73 in 0ab3499

DuplicateDynamicQuantChainPass()(model)

unfortunately there is a gap here for smooth quantization in dynamic, which makes this step manual, but we are working on resolving it.

[gpchowdari]

@gpchowdari if you're using dynamic quantization you might have to use the following pass:

executorch/examples/models/phi-3-mini/export_phi-3-mini.py

Line 73 in 0ab3499

DuplicateDynamicQuantChainPass()(model)
unfortunately there is a gap here for smooth quantization in dynamic, which makes this step manual, but we are working on resolving it.

@mcr229 with the suggestion, I am able to export the model. Thank you :)
Thanks @mergennachin @tarun292 for all the support.

Inference is failing with

Error in XNNPACK: failed to define Fully Connected operator with input ID #2, filter ID #3, bias ID #4, and output ID #1: mismatching datatypes across input (QDINT8), filter (QINT8), bias (FP32), and output (FP32) (xnn_define_fully_connected, executorch/backends/xnnpack/third-party/XNNPACK/src/subgraph/fully-connected.c:1216)
E 00:00:01.003792 executorch:XNNCompiler.cpp:750] Failed to create linear node 1381, with code: xnn_status_invalid_parameter
E 00:00:01.003801 executorch:XNNPACKBackend.cpp:105] XNNCompiler::compileModel failed: 0x1
E 00:00:01.003804 executorch:method.cpp:109] Init failed for backend XnnpackBackend: 0x1
F 00:00:01.004876 executorch:executor_runner.cpp:224] In function main(), assert failed (method.ok()): Loading of method forward failed with status 0x1
Aborted

[gpchowdari]

Update:

With the below 2 changes inference is working without issues but very slow.

get_symmetric_quantization_config(is_per_channel=True,is_dynamic=True)

and

XnnpackDynamicallyQuantizedPartitioner instead of XnnpackPartitioner