Guidelines for getting entire files into context?

How To
1

I’m working with some larger files and find that both o1-mini and claude-3.5-sonnet tell me they only see parts of them, what they have is a “snippet” and tell me I need to give them the entire file - which I thought I already had.

One example I have is a 2126 line file that is 83Kb with coffee/javascript. When I ask questions about specific methods or parts of the code in a Composer interaction, I often get responses like:

While you’ve provided a partial snippet of overlay.coffee, here’s a high-level understanding based on the provided code: …

Are there guidelines or best practices for how to format, structure or limit the size of files to maximize the chance the models can parse and understand them?

2

you could try using a larger context model like Google’s gemini?

3

can you check how many tokens is the file you mentioned?

https://platform.openai.com/tokenizer

you could try splitting the docs too if you dont want to use gemini

4

For the file mentioned above with 2126 lines and 83KB, the tokenizer reports:

Tokens: 20,431
Characters: 82749

5

I think sonnet 3.5 should be able to handle that easily.

Not sure if Cursor only takes a small part even from a small file. That would be weird.