|
21 | 21 | "\n", |
22 | 22 | "This notebook performs SFT training and evaluation workflow on [OpenAI's GSM8K dataset](https://huggingface.co/datasets/openai/gsm8k).\n", |
23 | 23 | "The primary goal is to demonstrate the end-to-end process of:\n", |
24 | | - "1. Pre-SFT Evaluation: Calcuating baseline accuracy for the model before training.\n", |
| 24 | + "1. Pre-SFT Evaluation: Calculating baseline accuracy for the model before training.\n", |
25 | 25 | "2. SFT Training: Fine-tune the model using MaxText & Tunix SFT trainer.\n", |
26 | 26 | "3. Post-SFT Evaluation: Re-running the evaluation loop after training to measure the performance gain achieved by SFT.\n", |
27 | 27 | "\n", |
|
363 | 363 | " f\"Processing {NUM_TEST_SAMPLES} examples with a batch size of {BATCH_SIZE}. This will result in {TOTAL_BATCHES} total batches for the test run.\"\n", |
364 | 364 | " )\n", |
365 | 365 | "else:\n", |
366 | | - " print(\"Evaluation on test dataset is skipped. Plese set `RUN_EVALUATION=True` to run evaluation.\")" |
| 366 | + " print(\"Evaluation on test dataset is skipped. Please set `RUN_EVALUATION=True` to run evaluation.\")" |
367 | 367 | ] |
368 | 368 | }, |
369 | 369 | { |
|
419 | 419 | " ),\n", |
420 | 420 | " )\n", |
421 | 421 | "else:\n", |
422 | | - " print(\"Evaluation on test dataset is skipped. Plese set `RUN_EVALUATION=True` to run evaluation.\")" |
| 422 | + " print(\"Evaluation on test dataset is skipped. Please set `RUN_EVALUATION=True` to run evaluation.\")" |
423 | 423 | ] |
424 | 424 | }, |
425 | 425 | { |
|
443 | 443 | " print(\"Running Pre-SFT Evaluation...\")\n", |
444 | 444 | " score = evaluate_model(test_dataset, vllm_rollout, debug=False)\n", |
445 | 445 | "else:\n", |
446 | | - " print(\"Evaluation on test dataset is skipped. Plese set `RUN_EVALUATION=True` to run evaluation.\")" |
| 446 | + " print(\"Evaluation on test dataset is skipped. Please set `RUN_EVALUATION=True` to run evaluation.\")" |
447 | 447 | ] |
448 | 448 | }, |
449 | 449 | { |
|
464 | 464 | " f\"Percentage of test samples where the model's output adheres to the expected structure: {score['correct_format']}%\"\n", |
465 | 465 | " )\n", |
466 | 466 | "else:\n", |
467 | | - " print(\"Evaluation on test dataset is skipped. Plese set `RUN_EVALUATION=True` to run evaluation.\")" |
| 467 | + " print(\"Evaluation on test dataset is skipped. Please set `RUN_EVALUATION=True` to run evaluation.\")" |
468 | 468 | ] |
469 | 469 | }, |
470 | 470 | { |
|
515 | 515 | " vllm_rollout.update_params(state)\n", |
516 | 516 | " score = evaluate_model(test_dataset, vllm_rollout, debug=False)\n", |
517 | 517 | "else:\n", |
518 | | - " print(\"Evaluation on test dataset is skipped. Plese set `RUN_EVALUATION=True` to run evaluation.\")" |
| 518 | + " print(\"Evaluation on test dataset is skipped. Please set `RUN_EVALUATION=True` to run evaluation.\")" |
519 | 519 | ] |
520 | 520 | }, |
521 | 521 | { |
|
538 | 538 | " f\"Percentage of test samples where the model's output adheres to the expected structure: {score['correct_format']}%\"\n", |
539 | 539 | " )\n", |
540 | 540 | "else:\n", |
541 | | - " print(\"Evaluation on test dataset is skipped. Plese set `RUN_EVALUATION=True` to run evaluation.\")" |
| 541 | + " print(\"Evaluation on test dataset is skipped. Please set `RUN_EVALUATION=True` to run evaluation.\")" |
542 | 542 | ] |
543 | 543 | } |
544 | 544 | ], |
|
0 commit comments