Fix Cell 16: remove tokenizer arg from Seq2SeqTrainer (renamed in transformers 5.x)

notebooks/kaggle_master_trainer.ipynb

@@ -171,7 +171,7 @@
    "id": "cell-train",
    "metadata": {},
    "outputs": [],
-   "source": "# ── Cell 16: TRAIN — Deep Sleep ───────────────────────────────────────────────\n# This is the long-running cell.  Expected wall time on T4:\n#   2000 steps ≈ 25 min   |   4000 steps ≈ 50 min   |   8000 steps ≈ 100 min\n#\n# If the session times out, just re-run the notebook from Cell 1.\n# LAST_CHECKPOINT (Cell 6) will be detected and training resumes automatically.\n\nfrom transformers import Seq2SeqTrainer\n\ntrainer = Seq2SeqTrainer(\n    model=model,\n    args=training_args,\n    train_dataset=train_ds,\n    eval_dataset=eval_ds,\n    data_collator=collator,\n    compute_metrics=compute_metrics,\n    tokenizer=processor.feature_extractor,\n)\n\nprint(f'Starting Deep Sleep training ...')\nprint(f'  Resume from : {LAST_CHECKPOINT or \"scratch\"}')\nprint(f'  Train size  : {len(train_ds)}')\nprint(f'  Eval size   : {len(eval_ds)}')\nprint()\n\ntrain_result = trainer.train(resume_from_checkpoint=LAST_CHECKPOINT)\n\nprint('\\n✅ Training complete')\nprint(f'  Steps    : {train_result.global_step}')\nprint(f'  Train loss: {train_result.training_loss:.4f}')\n\n# Save the best model weights locally\ntrainer.save_model(OUTPUT_DIR)\nprocessor.save_pretrained(OUTPUT_DIR)\nprint(f'  Adapter saved → {OUTPUT_DIR}')"
+   "source": "# -- Cell 16: TRAIN -----------------------------------------------------------\nfrom transformers import Seq2SeqTrainer\n\ntrainer = Seq2SeqTrainer(\n    model=model,\n    args=training_args,\n    train_dataset=train_ds,\n    eval_dataset=eval_ds,\n    data_collator=collator,\n    compute_metrics=compute_metrics,\n    # 'tokenizer' argument removed -- renamed to 'processing_class' in\n    # transformers 5.x and passing the old name raises TypeError\n)\n\nprint(f'Starting training ...')\nprint(f'  Resume from : {LAST_CHECKPOINT or \"scratch\"}')\nprint(f'  Train size  : {len(train_ds)}')\nprint(f'  Eval size   : {len(eval_ds)}')\n\ntrain_result = trainer.train(resume_from_checkpoint=LAST_CHECKPOINT)\n\nprint('\\nTraining complete')\nprint(f'  Steps     : {train_result.global_step}')\nprint(f'  Train loss: {train_result.training_loss:.4f}')\n\ntrainer.save_model(OUTPUT_DIR)\nprocessor.save_pretrained(OUTPUT_DIR)\nprint(f'  Adapter saved -> {OUTPUT_DIR}')"
   },
   {
    "cell_type": "markdown",