Generic LLM Interface

[DavidKoleczek]

Create a new generic wrapper for LLMs that can easily be extended to other providers.

[Copilot]

Copilot reviewed 53 out of 67 changed files in this pull request and generated 1 comment.

Files not reviewed (14)

• .vscode/settings.json: Language not supported
• notebooks/local_llm/10_ollama_deepseek_coder_v2.ipynb: Evaluated as low risk
• src/not_again_ai/llm/gh_models/azure_ai_client.py: Evaluated as low risk
• notebooks/local_llm/20_phi-3-vision.ipynb: Evaluated as low risk
• notebooks/llm/01_openai_chat_completion.ipynb: Evaluated as low risk
• notebooks/llm/02_ollama_intro.ipynb: Evaluated as low risk
• notebooks/llm/10_gpt-4-v.ipynb: Evaluated as low risk
• notebooks/local_llm/01_common_chat_completion.ipynb: Evaluated as low risk
• notebooks/local_llm/02_ollama_intro.ipynb: Evaluated as low risk
• README.md: Evaluated as low risk
• .github/workflows/python.yml: Evaluated as low risk
• pyproject.toml: Evaluated as low risk
• src/not_again_ai/llm/chat_completion/init.py: Evaluated as low risk
• src/not_again_ai/llm/chat_completion/interface.py: Evaluated as low risk

Comments suppressed due to low confidence (6)

src/not_again_ai/llm/chat_completion/providers/openai_api.py:70

• The code converts tool_call['function']['arguments'] to a string, but it does not handle cases where arguments might already be a string. This could lead to double stringification.

tool_call['function']['arguments'] = str(tool_call['function']['arguments'])

src/not_again_ai/llm/chat_completion/providers/openai_api.py:125

• The code attempts to load message.get('content', '{}') as JSON, but it does not handle cases where content is not a valid JSON string.

json_message = json.loads(message.get('content', '{}'))

src/not_again_ai/llm/chat_completion/providers/openai_api.py:220

• The openai_client function has multiple branches based on api_type, api_key, and other parameters. Ensure that all branches are covered by tests.

def openai_client(api_type: Literal['openai', 'azure_openai'] = 'openai',

src/not_again_ai/llm/chat_completion/providers/ollama_api.py:67

• The parameter 'json_mode' is used inconsistently. Explicitly check for True or False to avoid unexpected behavior.

json_mode = kwargs.get("json_mode", None)

src/not_again_ai/llm/chat_completion/types.py:109

• The stop field can be either a string or a list of strings, which could lead to ambiguity. Consider using a single type for consistency.

stop: str | list[str] | None = Field(default=None)

src/not_again_ai/llm/chat_completion/types.py:126

• The json_message field is defined as dict[str, Any] | None. Consider using a more specific type if possible.

json_message: dict[str, Any] | None = Field(default=None)