mirror of
https://github.com/saymrwulf/onnxruntime.git
synced 2026-07-25 19:48:11 +00:00
feeb0b50f9
### Description This PR improves `TreeNodeElementId` hash function by employing [Elegant Pairing function](http://szudzik.com/ElegantPairing.pdf). In few works, Elegant Pairing function maps two non−negative integers to a non−negative integer that is uniquely associated with that pair. This drastically reduces the collision and therefore reduces the time required to create a session in order to use a large tree ensemble model. ### Motivation and Context We use ONNX runtime to serve our models as part of Triton backend. We noticed that it was taking around 2 minutes to load a model which is a large tree ensemble model (around 5k trees with around 3 millions nodes in total). After investigating the issue, it was clear that the `TreeNodeElementId` hash function wasn't being able to map keys to buckets of C++ `unordered_map` without a significant amount of collisions (in same cases 700 items per bucket). The following picture shows graphically the improvement obtained by the proposed change. We used the `onnx_test_runner` command.  #### Before ``` $> time ./onnx_test_runner -v ~/folder_with_model result: Models: 1 Total test cases: 0 Succeeded: 0 Not implemented: 0 Failed: 0 Stats by Operator type: Not implemented(0): Failed: Failed Test Cases: real 0m55.695s user 0m52.919s sys 0m0.760s ``` #### After ``` $> time ./onnx_test_runner -v ~/folder_with_model result: Models: 1 Total test cases: 0 Succeeded: 0 Not implemented: 0 Failed: 0 Stats by Operator type: Not implemented(0): Failed: Failed Test Cases: real 0m17.152s user 0m14.318s sys 0m0.619s ``` |
||
|---|---|---|
| .. | ||
| codegen | ||
| common | ||
| dll | ||
| dlpack | ||
| eager | ||
| flatbuffers | ||
| framework | ||
| graph | ||
| language_interop_ops | ||
| mlas | ||
| optimizer | ||
| platform | ||
| providers | ||
| quantization | ||
| session | ||
| util | ||