I have a usage issue, but I'm not quite sure if it's a BUG

[luoyq-Rockey]

my gitlab merge request compare like this:

Then I triggered the webhook in the comment section by using /imporve
The results are as follows

As you can see, all the suggestions come from the directory "error_demo/", but actually I do have the directory "error_demo2/", and this directory also contains incorrect function methods

In the configuration.toml configuration file, I changed both num_code_suggestions_per_chunk and max_number_of_calls to 9

How can I overwrite all diffs when triggering CR?

[luoyq-Rockey]

@mrT23 Excuse me. Could you offer some help? Thank you very much

[luoyq-Rockey]

I attempted to adjust the value of max_model_tokens in configuration.toml from 32,000 to 64,000. I found that the suggestions included the error_demo2 section. Judging from the phenomenon, it seems that the tokens in my diff content are too long. However, the value of max_model_tokens is set too small

[IsmaelMartinez]

This looks like expected behaviour rather than a bug. num_code_suggestions_per_chunk is an upper bound per chunk, and for an MR small enough to fit the model context the whole diff is one chunk, so raising it and max_number_of_calls to 9 has no effect. Coverage is a separate matter: files with filtered or unrecognised extensions are dropped or deprioritised before the model sees them, so please check what extensions the files under error_demo2/ use. Running with verbosity_level=2 will log which files were treated as invalid. If suggestions are simply clustering in one file, set pr_code_suggestions.max_suggestions_per_file to a positive integer, which caps retention per file after chunks are merged and leaves room for the others. Closing for staleness, and happy to reopen with the file list and a debug log.