Please support deepseek coder greater-ly.

[sprappcom]

Issue

deepseek is much cheaper and useful for certain scenario but i seemed to have hit some limits:

Model deepseek/deepseek-coder has hit a token limit!
Token counts below are approximate.

Input tokens: ~18,596 of 128,000
Output tokens: ~3,223 of 4,096 -- possibly exceeded output limit!
Total tokens: ~21,819 of 128,000

For more info: https://aider.chat/docs/troubleshooting/token-limits.html

please make it "smarter" by allowing us to break down it's output generation in piecemeals or something.

how to make this work?

split the router.rs file into smaller files for easier management. (do for the first 2 files first)

optionally i did "do the first 2 files first" when the initial prompt didnt work but it still shows token limit hit etc.

pls help suggest ways to work around. thx

Version and model info

aider latest ubuntu 24.04

[olafgeibig]

DeepSeek now offers a beta API with 8k output tokens. This might ease the pain. https://platform.deepseek.com/api-docs/pricing/

[sprappcom]

@olafgeibig how do i do the "beta" that offers the 8k version?

p.s. : does it mean we shld request support for a new /model ...?

[paul-gauthier]

Thanks for trying aider and filing this issue.

I'm waiting on litellm support for prefill with deepseek.

https://github.com/BerriAI/litellm/issues/4881

[mr-september]

Are the 8K tokens and the prefill functionality actually referring to the same thing? Or would we benefit from using both prefill and the 8K api endpoint instead of just one of them?

[paul-gauthier]

The main branch has 3 new features which I think solve this issue:

• Infinite output for DeepSeek Coder, Mistral models in addition to Anthropic's models.
• New --deepseek switch to use DeepSeek Coder.
• DeepSeek Coder uses 8k token output.

The change is available in the main branch. You can get it by installing the latest version from github:

python -m pip install --upgrade git+https://github.com/paul-gauthier/aider.git

If you have a chance to try it, let me know if it works well for you.

[mr-september]

Is this right?

python -m aider.main --model deepseek --browser

It runs but I get this error with any prompt

BadRequestError: litellm.BadRequestError: GetLLMProvider Exception - list index out of range

original model: deepseek

The old way I ran still works fine:

python -m aider.main --model deepseek/deepseek-coder --browser

[paul-gauthier]

You need to use --model deepseek/deepseek-coder.

I'm going to close this issue for now, but feel free to add a comment here and I will re-open or file a new issue any time.