server
server copied to clipboard
run testing triton fail
libtorch_int32_int32_int32 | 1 | UNAVAILABLE: Internal: input 'INPUT0' does not follow naming convention i.e. <name>__<index>. |
| libtorch_int32_int32_int32 | 2 | UNAVAILABLE: Internal: input 'INPUT0' does not follow naming convention i.e. <name>__<index>. |
| libtorch_int32_int32_int32 | 3 | UNAVAILABLE: Internal: input 'INPUT0' does not follow naming convention i.e. <name>__<index>. |
when i run the ci test, i get this error, because in the config.pbtxt, the name of model's input is INPUT0, not INPUT__0, so it fails,but i follow the instructions to generate ci test as below: $ cd qa/common $ ./gen_qa_model_repository $ ./gen_qa_custom_ops https://github.com/triton-inference-server/server/blob/main/docs/test.md can someone tell me why are the config and the code is not consistent?thanks
Hi @LucQueen, can you clarify which test you are running? You can read more about the PyTorch naming conventions here: https://github.com/triton-inference-server/server/blob/main/docs/model_configuration.md#special-conventions-for-pytorch-backend
What is the version of Torch backend that you used? We recently added changes to relax the convention restriction as posted above and I believe the new test models are generated without the convention.
Closing issue due to lack of activity. Please re-open the issue if you would like to follow up with this.