mirror of
https://github.com/Aider-AI/aider.git
synced 2025-05-28 08:14:59 +00:00
Added improved editing support for command-r-plus
This commit is contained in:
parent
b882d5db4c
commit
cd838901d1
4 changed files with 34 additions and 5 deletions
15
docs/llms.md
15
docs/llms.md
|
@ -4,9 +4,11 @@
|
|||
Aider works well with OpenAI's GPT 3.5, GPT-4, GPT-4 Turbo with Vision and
|
||||
Anthropic's Claude 3 Opus and Sonnet.
|
||||
|
||||
GPT-4 Turbo and Claude 3 Opus are recommended for the best results.
|
||||
GPT-4 Turbo and Claude 3 Opus are recommended, as they are the very best coding assistants.
|
||||
Cohere offers *free* API access to their Command-R+ model, which works well with aider
|
||||
as a *very basic* coding assistant.
|
||||
|
||||
Aider also supports connecting to almost any LLM,
|
||||
Aider supports connecting to almost any LLM,
|
||||
but it may not work well with some models depending on their capabilities.
|
||||
For example, GPT-3.5 is just barely capable of reliably *editing code* to provide aider's
|
||||
interactive "pair programming" style workflow.
|
||||
|
@ -14,6 +16,7 @@ So you should expect that models which are less capable than GPT-3.5 may struggl
|
|||
|
||||
- [OpenAI](#openai)
|
||||
- [Anthropic](#anthropic)
|
||||
- [Cohere](#cohere)
|
||||
- [Azure](#azure)
|
||||
- [OpenAI compatible APIs](#openai-compatible-apis)
|
||||
- [Other LLMs](#other-llms)
|
||||
|
@ -54,6 +57,14 @@ You can use `aider --model <model-name>` to use any other Anthropic model.
|
|||
For example, if you want to use a specific version of Opus
|
||||
you could do `aider --model claude-3-opus-20240229`.
|
||||
|
||||
## Cohere
|
||||
|
||||
To work with Cohere's models, you need to provide your
|
||||
[Cohere API key](https://dashboard.cohere.com/)
|
||||
in the `COHERE_API_KEY` environment variable.
|
||||
|
||||
- **Command-R+**: `aider --model command-r-plus`
|
||||
|
||||
## Azure
|
||||
|
||||
Aider can be configured to connect to the OpenAI models on Azure.
|
||||
|
|
Loading…
Add table
Add a link
Reference in a new issue