fixes for code

[d-kleine]

• updated .gitignore for ch07/01
• removed unused import in ch04/01_main-chapter-code/exercise-solutions.ipynb (GELU() is in FeedForward() func)
• fixed code in ch05/02_alternative_weight_loading/weight-loading-hf-transformers.ipynb
• removed unused import in ch04/02_performance-analysis/flops-analysis.ipynb (tiktoken not used)

Please also see below comments, there are some issues that you might need to take a look into

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[d-kleine]

Please see

ch05/05_bonus_hparam_tuning/hparam_search.py

https://github.com/rasbt/LLMs-from-scratch/blob/1a65020d817afb1a2c2eb1ca5588b265ab055258/ch05/05_bonus_hparam_tuning/hparam_search.py#L53C1-L59C32

calc_loss_loader() in evaluate_model() uses keyword arg 'num_iters', even though it's 'num_batches'
https://github.com/rasbt/LLMs-from-scratch/blob/1a65020d817afb1a2c2eb1ca5588b265ab055258/ch05/05_bonus_hparam_tuning/hparam_search.py#L27C1-L41C36

[d-kleine]

ch05/05_bonus_hparam_tuning/previous_chapters.py

LLMs-from-scratch/ch05/05_bonus_hparam_tuning/previous_chapters.py

Line 48 in 1a65020

dataset = GPTDatasetV1(txt, tokenizer, max_length, stride, num_workers=0)

Unexpected keyword argument 'num_workers' in constructor call, not defined in

class GPTDatasetV1(Dataset):
    def __init__(self, txt, tokenizer, max_length, stride):

[d-kleine]

appendix-E/01_main-chapter-code/previous_chapters.py

https://github.com/rasbt/LLMs-from-scratch/blob/1a65020d817afb1a2c2eb1ca5588b265ab055258/appendix-E/01_main-chapter-code/previous_chapters.py#L486C1-L490C39

keyword arg 'tokenizer' is unused in train_classifier_simple().

[d-kleine]

GitHub Actions has encountered an internal error when running your job.
-> idk why, but the CI checks failed. I cannot run them again, can you please do that?

[rasbt]

Thanks for the PR. And the CI failure is weird, I'll try to restart.

[rasbt]

Looks really good to me, thanks a lot!

[rasbt]

I think the hparam num_iter was simply outdated. Must have forgotten to rename it there when I updated the chapter. Same for the num_workers.

Btw. great catch regarding the redundant tokenizer argument in the train_classifier_simple function. I think I used that originally in the "not _simple version but must have forgotten to remove it.

[d-kleine]

Thanks a lot for checking and updating the code!