[GitHub] [tvm] mkroening opened a new issue #9656: [Bug] TVMC tune: Broadcast exception on valid ONNX model

[GitBox <gi...@apache.org>]

mkroening opened a new issue #9656:
URL: https://github.com/apache/tvm/issues/9656


   ### Expected behavior
   
   No issue when running
   
   ```console
   tvmc tune --target llvm --output autotuner_records.json model.onnx
   ```
   
   with the ONNX model only being a multiplication of two four by four matrices (three by three works).
   Note that compiling the model without tuning works and produces valid results.
   
   Model:
   
   ```onnx
   ir_version: 8
   graph {
     node {
       input: "A"
       input: "B"
       output: "Y"
       op_type: "Gemm"
     }
     name: "test-model"
     input {
       name: "A"
       type {
         tensor_type {
           elem_type: 1
           shape {
             dim {
               dim_value: 4
             }
             dim {
               dim_value: 4
             }
           }
         }
       }
     }
     input {
       name: "B"
       type {
         tensor_type {
           elem_type: 1
           shape {
             dim {
               dim_value: 4
             }
             dim {
               dim_value: 4
             }
           }
         }
       }
     }
     output {
       name: "Y"
       type {
         tensor_type {
           elem_type: 1
           shape {
             dim {
               dim_value: 4
             }
             dim {
               dim_value: 4
             }
           }
         }
       }
     }
   }
   opset_import {
     version: 15
   }
   ```
   
   ### Actual behavior
   
   ```
   # [..]
   Traceback (most recent call last):
     File "/usr/lib/python3.8/runpy.py", line 194, in _run_module_as_main
       return _run_code(code, main_globals, None,
     File "/usr/lib/python3.8/runpy.py", line 87, in _run_code
       exec(code, run_globals)
     File "/home/mkroening/Development/tvm/python/tvm/driver/tvmc/__main__.py", line 24, in <module>
       tvmc.main.main()
     File "/home/mkroening/Development/tvm/python/tvm/driver/tvmc/main.py", line 100, in main
       sys.exit(_main(sys.argv[1:]))
     File "/home/mkroening/Development/tvm/python/tvm/driver/tvmc/main.py", line 93, in _main
       return args.func(args)
     File "/home/mkroening/Development/tvm/python/tvm/driver/tvmc/autotuner.py", line 266, in drive_tune
       tune_model(
     File "/home/mkroening/Development/tvm/python/tvm/driver/tvmc/autotuner.py", line 486, in tune_model
       tune_tasks(tasks, tuning_records, **tuning_options)
     File "/home/mkroening/Development/tvm/python/tvm/driver/tvmc/autotuner.py", line 687, in tune_tasks
       tuner_obj.tune(
     File "/home/mkroening/Development/tvm/python/tvm/autotvm/tuner/xgboost_tuner.py", line 105, in tune
       super(XGBTuner, self).tune(*args, **kwargs)
     File "/home/mkroening/Development/tvm/python/tvm/autotvm/tuner/tuner.py", line 169, in tune
       self.update(inputs, results)
     File "/home/mkroening/Development/tvm/python/tvm/autotvm/tuner/model_based_tuner.py", line 281, in update
       self.cost_model.fit(self.xs, self.ys, self.plan_size)
     File "/home/mkroening/Development/tvm/python/tvm/autotvm/tuner/xgboost_cost_model.py", line 179, in fit
       x_train = self._get_feature(xs)
     File "/home/mkroening/Development/tvm/python/tvm/autotvm/tuner/xgboost_cost_model.py", line 341, in _get_feature
       ret[i, :] = t if t is not None else 0
   ValueError: could not broadcast input array from shape (468,) into shape (426,)
    Done.
   ```
   
   ### Environment
   
   Operating System: Ubuntu 20.04.3 LTS
   TVM version: ccd59e89d21cc81cc06f2a16cddcc1ffeed1e2a1
   
   ### Steps to reproduce
   
   Create `model.onnx` with:
   
   ```python
   import onnx
   from onnx import helper
   from onnx import TensorProto
   
   l = 4
   m = 4
   n = 4
   
   A = helper.make_tensor_value_info('A', TensorProto.FLOAT, [l, m])
   B = helper.make_tensor_value_info('B', TensorProto.FLOAT, [m, n])
   
   Y = helper.make_tensor_value_info('Y', TensorProto.FLOAT, [l, n])
   
   node_def = helper.make_node(
       'Gemm',         # name
       ['A', 'B'],     # inputs
       ['Y'],          # outputs
   )
   
   graph_def = helper.make_graph(
       [node_def],     # nodes
       'test-model',   # name
       [A, B],         # inputs
       [Y],            # outputs
   )
   
   model_def = helper.make_model(graph_def)
   
   print('The model is:\n{}'.format(model_def))
   onnx.checker.check_model(model_def)
   print('The model is checked!')
   
   onnx.save(model_def, 'model.onnx')
   ```
   
   Thanks a lot for your help! :)
   


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org