Spaces:
Runtime error
Runtime error
Commit History
-
Ran 3rd 1 epoch, now ran 3 epochs at once (6 epochs trained now)
stefhooye28aef2
-
Ran a 2nd epoch
stefhooyf054892
-
First run has been completed
stefhooypython -m src.train --epochs 1 --batch-size 8: Loading data... IBM Debater: 34 debates CMV file missing — loading ConvoKit corpus… CMV (ConvoKit): 2982 debates (total: 3016) Device: cuda 50395 total examples Some weights of RobertaForSequenceClassification were not initialized from the model checkpoint at roberta-base and are newly initialized: ['classifier.dense.bias', 'classifier.dense.weight', 'classifier.out_proj.bias', 'classifier.out_proj.weight'] You should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference. Epoch 1/1: 100%|███████████████████████████████████████████████████████████████| 5355/5355 [56:47<00:00, 1.57it/s] Epoch 1: train_loss=0.6114 val_loss=0.5064 val_acc=0.7742 val_f1=0.8093 → Saved best checkpoint (F1=0.8093) to models\best/ Training complete. Best val F1: 0.8093e1f2faf
-
Created src/dataset.py — converts Debate objects into PyTorch tensors for RoBERTa
stefhooyCreated src/model.py — loads roberta-base configured for 4-class classification Created src/train.py — full training loop with class weights, AdamW optimizer, validation, checkpoint saving Created src/__init__.py — predict() and predict_debate() functions for other team members to use Fixed Person 1's IBM loader bug (was ignoring stance_WA, labeling everything as premise) Trained a first model (wrong labels, now discarded) What still needs to be done right now: Retrain with the fixed labels — run python -m src.train --epochs 3 --batch-size 8 Sanity check the model once training finishes — run a quick predict test to confirm it returns different labels for different inputs Commit and push the fixed code to the repo so the rest of the team can use predict() and predict_debate()deefbae