scripts/transformers/run_fewshot.py [55:90]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
    def model_init():
        return AutoModelForSequenceClassification.from_pretrained(
            model_id, num_labels=num_labels, id2label=id2label, label2id=label2id
        )

    # Define metrics
    metric_fn = load(metric)

    def compute_metrics(pred):
        labels = pred.label_ids
        preds = pred.predictions.argmax(-1)
        return metric_fn.compute(predictions=preds, references=labels)

    for idx, (split, dset) in enumerate(fewshot_dset.items()):
        typer.echo(f"🍌🍌🍌 Fine-tuning on {dataset_id} with split: {split} 🍌🍌🍌")
        # Create split directory
        metrics_split_dir = metrics_dir / split
        metrics_split_dir.mkdir(parents=True, exist_ok=True)
        metrics_filepath = metrics_split_dir / "results.json"
        # Skip previously evaluated split
        if metrics_filepath.is_file():
            typer.echo(f"INFO -- split {split} already trained, skipping ...")
            continue

        if debug:
            if split.split("-")[1] in ["4", "8", "16", "32", "64"]:
                break
            if idx > 0:
                break

        # Create training and validation splits
        dset = dset.train_test_split(seed=42, test_size=0.2)

        # Define hyperparameters
        ckpt_name = f"{model_name}-finetuned-{dataset_id}-{split}"
        training_args = TrainingArguments(
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -



scripts/transformers/run_fewshot_multilingual.py [80:115]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
    def model_init():
        return AutoModelForSequenceClassification.from_pretrained(
            model_id, num_labels=num_labels, id2label=id2label, label2id=label2id
        )

    # Define metrics
    metric_fn = load(metric)

    def compute_metrics(pred):
        labels = pred.label_ids
        preds = pred.predictions.argmax(-1)
        return metric_fn.compute(predictions=preds, references=labels)

    for idx, (split, dset) in enumerate(fewshot_dset.items()):
        typer.echo(f"🍌🍌🍌 Fine-tuning on {dataset_id} with split: {split} 🍌🍌🍌")
        # Create split directory
        metrics_split_dir = metrics_dir / split
        metrics_split_dir.mkdir(parents=True, exist_ok=True)
        metrics_filepath = metrics_split_dir / "results.json"
        # Skip previously evaluated split
        if metrics_filepath.is_file():
            typer.echo(f"INFO -- split {split} already trained, skipping ...")
            continue

        if debug:
            if split.split("-")[1] in ["4", "8", "16", "32", "64"]:
                break
            if idx > 0:
                break

        # Create training and validation splits
        dset = dset.train_test_split(seed=42, test_size=0.2)

        # Define hyperparameters
        ckpt_name = f"{model_name}-finetuned-{dataset_id}-{split}"
        training_args = TrainingArguments(
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -