Remove bad eval when onnx session used

[msaroufim]

Fixes #2033

I believe this error occured when I tried to merge the dynamo branch with the onnx branch since this line is no longer needed since eval happens directly if model is eager mode or ts

If the ONNX test was enabled by default we would have caught this when merging the torch.compile PR.

In a follow-up PR I'm going to

• refactor the base handler a bit more
• Make all the optimization-related dependencies mandatory
• Make our requirements.txt more modular
• Get back to doing releases with our non dev docker binaries

Tests

pip install onnx
(serve) ubuntu@ip-172-31-17-70:~/serve/test/pytest$ pytest test_onnx.py 
======================================================================== test session starts =========================================================================
platform linux -- Python 3.8.13, pytest-7.2.0, pluggy-1.0.0
rootdir: /home/ubuntu/serve
collected 5 items                                                                                                                                                    

test_onnx.py .....                                                                                                                                             [100%]

========================================================================= 5 passed in 2.08s ==========================================================================

torch.compile tests

(base) ubuntu@ip-172-31-17-70:~$ curl http://127.0.0.1:8080/predictions/densenet161 -T kitten_small.jpg

{
  "Siberian_husky": 0.006150901783257723,
  "grille": 0.005219979677349329,
  "submarine": 0.004688442684710026,
  "puck": 0.004634885583072901,
  "mountain_tent": 0.004277920816093683

2022-12-13T19:14:28,584 [INFO ] W-9000-densenet161_1.0-stdout MODEL_LOG - Listening on port: /tmp/.ts.sock.9000
2022-12-13T19:14:28,589 [INFO ] W-9000-densenet161_1.0-stdout MODEL_LOG - Successfully loaded /home/ubuntu/serve/ts/configs/metrics.yaml.
2022-12-13T19:14:28,590 [INFO ] W-9000-densenet161_1.0-stdout MODEL_LOG - [PID]2795
2022-12-13T19:14:28,590 [INFO ] W-9000-densenet161_1.0-stdout MODEL_LOG - Torch worker started.
2022-12-13T19:14:28,591 [INFO ] W-9000-densenet161_1.0-stdout MODEL_LOG - Python runtime: 3.8.13
2022-12-13T19:14:28,608 [INFO ] W-9000-densenet161_1.0-stdout MODEL_LOG - Connection accepted: /tmp/.ts.sock.9000.
2022-12-13T19:14:28,649 [INFO ] W-9000-densenet161_1.0-stdout MODEL_LOG - model_name: densenet161, batchSize: 1
2022-12-13T19:14:36,533 [INFO ] W-9000-densenet161_1.0-stdout MODEL_LOG - Compiled model with backend inductor
2022-12-13T19:14:49,290 [INFO ] W-9000-densenet161_1.0-stdout MODEL_LOG - Backend received inference at: 1670958889

[codecov]

Codecov Report

Merging #2034 (8176c72) into master (3a0f8f8) will increase coverage by 0.01%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master    #2034      +/-   ##
==========================================
+ Coverage   53.36%   53.38%   +0.01%     
==========================================
  Files          71       71              
  Lines        3225     3224       -1     
  Branches       56       56              
==========================================
  Hits         1721     1721              
+ Misses       1504     1503       -1     

Impacted Files	Coverage Δ
ts/torch_handler/base_handler.py	0.00% <ø> (ø)

📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more