onnx model dynamic axis not working

[wooAo]

Description onnx model dynamic axis is not working i'm sure the onnx output node has the dynamic axis for both dimension

netron info:

export script:

print onnx model output node info:

[name: "output"
type {
  tensor_type {
    elem_type: 1
    shape {
      dim {
        dim_param: "None"
      }
      dim {
        dim_param: "None"
      }
    }
  }
}
]

Triton Information model_repository_manager.cc:832] failed to load 'Drinks-Sku' version 1: Invalid argument: model 'Drinks-Sku', tensor 'output': the model expects 2 dimensions (shape [-1,16075]) but the model configuration specifies 2 dimensions (an initial batch dimension because max_batch_size > 0 followed by the explicit tensor shape, making complete shape [-1,-1])

Are you using the Triton container or did you build it yourself? Triton container

config.pbtxt

name: "Drinks-Sku"
platform: "onnxruntime_onnx"
max_batch_size: 128
input [
  {
    name: "img"
    data_type: TYPE_FP32
    format: FORMAT_NHWC
    dims: [3, 224, 224]
  }
]
output [
  {
    name: "output"
    data_type: TYPE_FP32
    dims: [-1 ]
  }
]
instance_group [
  {
    count: 1
    kind: KIND_GPU
    gpus: [ 0 ]
  }
]
dynamic_batching { }

[rmccorm4]

Hi @wooAo ,

model_repository_manager.cc:832] failed to load 'Drinks-Sku' version 1: Invalid argument: model 'Drinks-Sku', tensor 'output': the model expects 2 dimensions (shape [-1,16075]) but the model configuration specifies 2 dimensions (an initial batch dimension because max_batch_size > 0 followed by the explicit tensor shape, making complete shape [-1,-1])

Is your output tensor supposed to be something like this instead in the config?

output [
  {
    name: "output"
    data_type: TYPE_FP32
    dims: [ 16075 ]   // implicitly [-1, 16075] because max_batch_size > 0
  }
]

[wooAo]

Hi, the specified dims work fine, but we have more than 10 models in our production line and they need to share one config.pbtxt (must be)

[nv-kmcgill53]

Are your 10 models different versions of the same one? If this is the case then you can layout your directory structure as follows:

<model-repository-path>/
  <model-name>/
    config.pbtxt
    1/
      onnx.model
    2/
      onnx.model
    3/
      onnx.model
    ...

If you want distinct names for your model then you will have to copy your config.pbtxt file (changing the name parameter appropriately) to each of your directories.

[krishung5]

Closing issue due to lack of activity. Please re-open the issue if you would like to follow up with this issue.