Issue
I am using aider to work on a "larger" git repository.
I noticed very high input token usage when sending queries to claude-3.5-sonnet.
Further investigations lead me to believe that aider is currently not respecting the repository map token limit.
Consider the below snippet, in which i call aider explicitly with "--map-tokens 1024", however the "/tokens" command still reports it to use over 16.000 tokens for the repository map.
The value reported is consistent with the API usage reports i get from the Anthropic API console.
I tried deleting the ".aider.tags.cache.v3" directory within the repository folder.
This lets aider regenerate the repository map cache, however the token usage stays the same.
Setting the map-tokens config key in the ".aider.conf.yml" file in my homedir also does not change the behavior.
Another symptom possibly related is the very slow autocomplete when adding files to the chat. Sometimes it takes several seconds before the list with the suggestions is generated.
PS C:\Data\repos\MyLargeProject> aider --map-tokens 1024
Aider v0.40.6
Models: claude-3-5-sonnet-20240620 with diff edit format, weak model claude-3-haiku-20240307
Git repo: .git with 552 files
Repo-map: using 1024 tokens
Use /help to see in-chat commands, run with --help to see cmd line args
────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────
> /tokens
Approximate context window usage, in tokens:
$ 0.0034 1,133 system messages
$ 0.0493 16,419 repository map use --map-tokens to resize
==================
$ 0.0527 17,552 tokens total
182,448 tokens remaining in context window
200,000 tokens max context window size
────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────
>
Version and model info
Aider v0.40.6
Models: claude-3-5-sonnet-20240620 with diff edit format, weak model claude-3-haiku-20240307
Git repo: .git with 552 files
Issue
I am using aider to work on a "larger" git repository.
I noticed very high input token usage when sending queries to claude-3.5-sonnet.
Further investigations lead me to believe that aider is currently not respecting the repository map token limit.
Consider the below snippet, in which i call aider explicitly with "--map-tokens 1024", however the "/tokens" command still reports it to use over 16.000 tokens for the repository map.
The value reported is consistent with the API usage reports i get from the Anthropic API console.
I tried deleting the ".aider.tags.cache.v3" directory within the repository folder.
This lets aider regenerate the repository map cache, however the token usage stays the same.
Setting the map-tokens config key in the ".aider.conf.yml" file in my homedir also does not change the behavior.
Another symptom possibly related is the very slow autocomplete when adding files to the chat. Sometimes it takes several seconds before the list with the suggestions is generated.
Version and model info
Aider v0.40.6
Models: claude-3-5-sonnet-20240620 with diff edit format, weak model claude-3-haiku-20240307
Git repo: .git with 552 files