Update-llama-code (#25826)

* some bug fixes

* updates

* Update code_llama.md

Co-authored-by: Omar Sanseviero <osanseviero@users.noreply.github.com>

* Add co author

Co-authored-by: pcuenca <pedro@latenitesoft.com>

* add a test

* fixup

* nits

* some updates

* fix-coies

* adress comments

* nits

* nits

* fix docsting

* Apply suggestions from code review

Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>

* update

* add int for https://huggingface.co/spaces/hf-accelerate/model-memory-usage

---------

Co-authored-by: Omar Sanseviero <osanseviero@users.noreply.github.com>
Co-authored-by: pcuenca <pedro@latenitesoft.com>
Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>

docs/source/en/model_doc/code_llama.md

@@ -49,6 +49,8 @@ Here is a sample usage
 python src/transformers/models/llama/convert_llama_weights_to_hf.py \
     --input_dir /path/to/downloaded/llama/weights --model_size 7B --output_dir /output/path
 ```
+Note that executing the script requires enough CPU RAM to host the whole model in float16 precision (even if the biggest versions
+come in several checkpoints they each contain a part of each weight of the model, so we need to load them all in RAM).
 
 - After conversion, the model and tokenizer can be loaded via:
 
@@ -90,8 +92,8 @@ If you only want the infilled part:
 >>> generator = pipeline("text-generation",model="codellama/CodeLlama-7b-hf",torch_dtype=torch.float16, device_map="auto")
 >>> generator('def remove_non_ascii(s: str) -> str:\n    """ <FILL_ME>\n    return result', max_new_tokens = 128, return_type = 1)
 ```
-Note that executing the script requires enough CPU RAM to host the whole model in float16 precision (even if the biggest versions
-come in several checkpoints they each contain a part of each weight of the model, so we need to load them all in RAM). For the 75B model, it's thus 145GB of RAM needed.
+
+Under the hood, the tokenizer [automatically splits by `<FILL_ME>`](https://huggingface.co/docs/transformers/main/model_doc/code_llama#transformers.CodeLlamaTokenizer.fill_token) to create a formatted input string that follows [the original training pattern](https://github.com/facebookresearch/codellama/blob/cb51c14ec761370ba2e2bc351374a79265d0465e/llama/generation.py#L402). This is more robust than preparing the pattern yourself: it avoids pitfalls, such as token glueing, that are very hard to debug.  To see how much CPU and GPU memory you need for this model or others, try [this calculator](https://huggingface.co/spaces/hf-accelerate/model-memory-usage) which can help determine that value.
 
 - The LLaMA tokenizer is a BPE model based on [sentencepiece](https://github.com/google/sentencepiece). One quirk of sentencepiece is that when decoding a sequence, if the first token is the start of the word (e.g. "Banana"), the tokenizer does not prepend the prefix space to the string.