Draft: rebase OGBG-code2

[puririshi98]

rebase #9652
Even with 10% of the data we can get F1>.5,

full e2e results get 3x SOTA F1 (sota=.22, ours=.62) (ours=CodeLLAMA+GAT)

as a baseline i tried using CodeLLAMA and LLAMA7b and both failed miserably to follow the task with any prompt i could try.
Will test finetuning CodeLLAMA next.

38899/50997 [5:50:15<1:39:25,  2.03it/s]
Epoch: 1|1: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 50997/50997 [7:39:38<00:00,  1.85it/s]
Epoch: 1|1, Train Loss: 0.577509
Final evaluation...
100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 1372/1372 [24:57<00:00,  1.06s/it]Hit: 0.6785
Precision: 0.6208
Recall: 0.6432
F1: 0.6201
Total Training Time: 29502.916083s
100%|███████████████████████████████

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 87.33%. Comparing base (ba3b906) to head (a84a2b6).
Report is 3 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #9671      +/-   ##
==========================================
- Coverage   87.54%   87.33%   -0.21%     
==========================================
  Files         482      482              
  Lines       31414    31411       -3     
==========================================
- Hits        27501    27434      -67     
- Misses       3913     3977      +64     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[Kh4L]

Few missing imports and arser.parse_args() but other than that LGTM!