File editing problems

Sometimes the LLM will reply with some code changes that don’t get applied to your local files. In these cases, aider might say something like “Failed to apply edit to filename” or other error messages.

This usually happens because the LLM is disobeying the system prompts and trying to make edits in a format that aider doesn’t expect. Aider makes every effort to get the LLM to conform, and works hard to deal with LLM edits that are “almost” correctly formatted.

But sometimes the LLM just won’t cooperate. In these cases, here are some things you might try.

Don’t add too many files

Many LLMs now have very large context windows, but filling them with irrelevant code or conversation can confuse the model.

  • Don’t add too many files to the chat, just add the files you think need to be edited. Aider also sends the LLM a map of your entire git repo, so other relevant code will be included automatically.
  • Use /drop to remove files from the chat session which aren’t needed for the task at hand. This will reduce distractions and may help the LLM produce properly formatted edits.
  • Use /clear to remove the conversation history, again to help the LLM focus.
  • Use /tokens to see how many tokens you are using for each message.

Use a more capable model

If possible try using GPT-4o, Claude 3.5 Sonnet or Claude 3 Opus, as they are the strongest and most capable models.

Weaker models are more prone to disobeying the system prompt instructions. Most local models are just barely capable of working with aider, so editing errors are probably unavoidable.

Local models: context window and quantization

Be especially careful about the Ollama context window when working with local models. It defaults to be very small and silently discards data if you exceed it.

Local models which have been quantized are more likely to have editing problems because they are not capable enough to follow aider’s system prompts.

Try the whole edit format

Run aider with --edit-format whole if were using a different edit format. You can see which edit format it is using in the announce lines:

Aider v0.50.2-dev
Models: claude-3-5-sonnet-20240620 with ♾️ diff edit format

More help

If you need more help, please check our GitHub issues and file a new issue if your problem isn’t discussed. Or drop into our Discord to chat with us.

When reporting problems, it is very helpful if you can provide:

  • Aider version
  • LLM model you are using

Including the “announcement” lines that aider prints at startup is an easy way to share this helpful info.

Aider v0.37.1-dev
Models: gpt-4o with diff edit format, weak model gpt-3.5-turbo
Git repo: .git with 243 files
Repo-map: using 1024 tokens

Use /help <question> to ask for help about using aider, customizing settings, troubleshooting, using LLMs, etc.