prepared for run by prof Filip Gralinski

code/README.md

@@ -20,9 +20,11 @@ pdm install
 ```
 ### 4. Run training code
 ```bash
-pdm run_training
+pdm train
 ```
-or
+
+## Required secrets
 ```
-pdm run src/train_codebert_mlm.py
+export HF_TOKEN=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
+export WANDB_API_KEY=313671f10f2a389b3171b32da8d4abdad91aaa7c
 ```

code/src/config.json

@@ -1,9 +1,9 @@
 {
-    "extra_embeddings": false,
-    "run_name": "original-continued",
+    "extra_embeddings": true,
+    "run_name": "tree-continued",
     "data_dir": "./data/codeparrot-clean-parsed-starencoder-no-comments/",
-    "output_dir": "./outputs/no-comments-starencoder-original-2",
-    "checkpoint": "./outputs/no-comments-starencoder-original/1_epoch_ckpt/",
+    "output_dir": "./outputs/no-comments-starencoder-tree-2",
+    "checkpoint": null,
     "seed": 420,
     "mlm_probability": 0.15,
     "batch_size": 32,

code/src/training.py

@@ -50,7 +50,7 @@ def main():
     set_seed(config['seed'])
 
     # Initialize W&B
-    wandb.init(project='codeparrot-starencoder-no-comments', config=config, name=config['run_name'])
+    wandb.init(project='gralinski', config=config, name=config['run_name'])
     
     # Upload the training files to W&B
     wandb.save(__file__)