Feature Request: Add DeepSeek Coder LLM as a Model Option

[jcraftsman]

Describe the feature

I would like to request the addition of DeepSeek Coder LLM as an available model option within SWE-agent. This integration would allow users to select DeepSeek Coder for their benchmarking and experimental needs, leveraging its capabilities without promoting it over other models.

Link to DeepSeek Coder: DeepSeek Coder GitHub Repository

Proposed Change:

Model Integration: Add DeepSeek Coder to the list of selectable models in SWE-agent.

Benefits:

Provides users with a cost-effective alternative model for experiments and benchmarking.
Enhances the flexibility and diversity of models available within SWE-agent.

Thank you for considering this feature request.

Potential Solutions

Potential Implementation Steps:

Update the SWE-agent configuration to include DeepSeek Coder as a model option.
In sweagent/agent/models.py:

"deepseek-coder": {
            "max_context": 32_000,
            "cost_per_input_token": 1.4e-07,
            "cost_per_output_token": 2.8e-07,
        },

#...
    elif (
        args.model_name.startswith("gpt")
        or args.model_name.startswith("ft:gpt")
        or args.model_name.startswith("azure:gpt")
        or args.model_name.startswith("deepseek")
    ):
        return OpenAIModel(args, commands)

I tested it as OpenAIModel and it worked.
Here is an example of keys.cfg:

OPENAI_API_BASE_URL: 'https://api.deepseek.com/v1'
OPENAI_API_KEY: 'sk-yourp9iva3k3y'