Add model name validation and constants for LLM providers
Description
Currently, when initializing the configuration classes for LLM providers (e.g. AnthropicConfig, OpenAIConfig, GoogleConfig), we pass model names as raw strings:
cfg = kit.summaries.AnthropicConfig(api_key=key, model="claude-3-haiku-20240307")
This approach has the following drawbacks (across all providers):
- No validation of model names
- Prone to typos and errors (e.g. "claude-3-haiku" vs. "claude-3-haiku-20240307")
- Hard to discover available models
Proposed Solution
- Create enum or constants for each LLM provider's models
- Add validation in
__post_init__ to check if the provided model name is valid
- Provide better developer experience with autocomplete and documentation
Example Implementation
from enum import Enum
class AnthropicModels(str, Enum):
CLAUDE_3_OPUS = "claude-3-opus-20240229"
CLAUDE_3_SONNET = "claude-3-sonnet-20240229"
...
class OpenAIModels(str, Enum):
GPT_4O = "gpt-4o"
GPT_4_TURBO = "gpt-4-turbo"
# Add other models as needed
class AnthropicConfig:
"""Configuration for Anthropic API access."""
api_key: Optional[str] = field(default_factory=lambda: os.environ.get("ANTHROPIC_API_KEY"))
model: str = AnthropicModels.CLAUDE_3_OPUS.value
temperature: float = 0.7
max_tokens: int = 1000
def __post_init__(self):
if not self.api_key:
raise ValueError(
"Anthropic API key not found. "
"Set ANTHROPIC_API_KEY environment variable or pass api_key directly."
)
# Validate model name
valid_models = [model.value for model in AnthropicModels]
if self.model not in valid_models:
raise ValueError(
f"Invalid model name for Anthropic: {self.model}. "
f"Please use one of: {', '.join(valid_models)}"
)
# Similar implementations for OpenAIConfig and GoogleConfig classesWith this implementation, users could initialize the config like:
# Using string (with validation)
cfg = kit.summaries.AnthropicConfig(api_key=key, model="claude-3-haiku-20240307")
# Or better, using constants
cfg = kit.summaries.AnthropicConfig(api_key=key, model=AnthropicModels.CLAUDE_3_HAIKU.value)
Benefits
- Type safety and validation across all LLM providers
- Better IDE support with autocomplete
- Clear documentation of available models
- Easier maintenance when adding/removing models
Add model name validation and constants for LLM providers
Description
Currently, when initializing the configuration classes for LLM providers (e.g.
AnthropicConfig,OpenAIConfig,GoogleConfig), we pass model names as raw strings:This approach has the following drawbacks (across all providers):
Proposed Solution
__post_init__to check if the provided model name is validExample Implementation
With this implementation, users could initialize the config like:
Benefits