Configure Aider with custom models?

I would like to use Aider with models hosted on glhf.chat.

I use this configuration:

~/aider.conf.yml

openai-api-base: https://glhf.chat/api/openai/v1
openai-api-key: glhf_MY_SECRET_API_KEY
model-settings-file: ~/.aider.model.settings.yml
model: deepseek-ai/DeepSeek-R1
weak-model: deepseek-ai/DeepSeek-V3
editor-model: Qwen/Qwen2.5-Coder-32B-Instruct

~/.aider.model.settings.yml

- name: deepseek-ai/DeepSeek-V3
  edit_format: diff
  use_repo_map: true
  reminder: sys
  examples_as_sys_msg: true
  extra_params:
    max_tokens: 8192
  caches_by_default: true

- name: deepseek-ai/DeepSeek-R1
  edit_format: diff
  weak_model_name: deepseek-ai/DeepSeek-V3
  use_repo_map: true
  examples_as_sys_msg: true
  extra_params:
    max_tokens: 8192
    include_reasoning: true
  caches_by_default: true
  editor_model_name: deepseek-ai/DeepSeek-V3
  editor_edit_format: editor-diff

- name: Qwen/Qwen2.5-Coder-32B-Instruct
  edit_format: diff
  weak_model_name: Qwen/Qwen2.5-Coder-32B-Instruct
  use_repo_map: true
  editor_model_name: Qwen/Qwen2.5-Coder-32B-Instruct
  editor_edit_format: editor-diff

But when I start aider it says:

───────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────
Warning for deepseek-ai/DeepSeek-R1: Unknown context window size and costs, using sane defaults.
Did you mean one of these?
- deepseek/deepseek-chat
- deepseek/deepseek-coder
Warning for deepseek-ai/DeepSeek-V3: Unknown context window size and costs, using sane defaults.
Did you mean one of these?
- deepseek/deepseek-chat
Warning for Qwen/Qwen2.5-Coder-32B-Instruct: Unknown context window size and costs, using sane defaults.
Did you mean one of these?
- openrouter/qwen/qwen-2.5-coder-32b-instruct
You can skip this check with --no-show-model-warnings

Why? How can I get rid of those warnings?

Billy (one of the glhf.chat founders) here! Thanks for using our app!

To remove the warnings, you can tell Aider about the models you're using with a .aider.model.metadata.json file. See Context window size and token costs

Here's one that should work for your models:

{
  "openai/hf:deepseek-ai/DeepSeek-R1": {
    "max_tokens": 8192,
    "max_input_tokens": 128000,
    "max_output_tokens": 8192,
    "input_cost_per_token": 0.000003,
    "output_cost_per_token": 0.000007,
    "mode": "chat",
  },
  "openai/hf:deepseek-ai/DeepSeek-V3": {
    "max_tokens": 8192,
    "max_input_tokens": 128000,
    "max_output_tokens": 8192,
    "input_cost_per_token": 0.00000125,
    "output_cost_per_token": 0.00000125,
    "mode": "chat",
  },
  "openai/hf:Qwen/Qwen2.5-Coder-32B-Instruct": {
    "max_tokens": 8192,
    "max_input_tokens": 32000,
    "max_output_tokens": 8192,
    "input_cost_per_token": 0.00000125,
    "output_cost_per_token": 0.00000125,
    "mode": "chat",
  },
}

(Aider also adds openai/ to specify OpenAI compatible providers so we also do that in the configs.)

Note that for glhf.chat's API, we need you need to add an hf: (HuggingFace) prefix for model names:

// .aider.model.settings.yml

- name: openai/hf:deepseek-ai/DeepSeek-V3
  edit_format: diff
  use_repo_map: true
  reminder: sys
  examples_as_sys_msg: true
  caches_by_default: true

- name: openai/hf:deepseek-ai/DeepSeek-R1
  edit_format: diff
  weak_model_name: openai/hf:deepseek-ai/DeepSeek-V3
  use_repo_map: true
  examples_as_sys_msg: true
  extra_params:
    include_reasoning: true
  caches_by_default: true
  editor_model_name: openai/hf:deepseek-ai/DeepSeek-V3
  editor_edit_format: editor-diff

- name: openai/hf:Qwen/Qwen2.5-Coder-32B-Instruct
  edit_format: diff
  weak_model_name: openai/hf:Qwen/Qwen2.5-Coder-32B-Instruct
  use_repo_map: true
  editor_model_name: openai/hf:Qwen/Qwen2.5-Coder-32B-Instruct
  editor_edit_format: editor-diff

// .aider.conf.yml

openai-api-base: https://api.glhf.chat/v1
openai-api-key: glhf_MY_SECRET_API_KEY
model-settings-file: ~/.aider.model.settings.yml
model: openai/hf:deepseek-ai/DeepSeek-R1
weak-model: openai/hf:deepseek-ai/DeepSeek-V3
editor-model: openai/hf:Qwen/Qwen2.5-Coder-32B-Instruct

Hope that helps!