prompttrail.models package

Submodules

prompttrail.models.anthropic module

pydantic model prompttrail.models.anthropic.AnthropicConfig

Bases: Config

Integration configuration class for Anthropic Claude API.

Manages authentication credentials and model parameters in a centralized way.

Fields:

• api_key (str)

• max_tokens (int | None)

• model_name (str)

• temperature (float | None)

• top_k (int | None)

• top_p (float | None)

Validators:

field api_key: str [Required]

API key for Anthropic API.

Validated by:

• validate_providers

field max_tokens: int | None = 1024

Maximum number of tokens to generate.

Validated by:

• validate_providers

field model_name: str = 'claude-3-opus-latest'

Name of the model to use.

Validated by:

• validate_providers

field temperature: float | None = 1.0

Temperature for sampling.

Validated by:

• validate_providers

field top_k: int | None = None

Top-k value for sampling.

Validated by:

• validate_providers

field top_p: float | None = None

Top-p value for sampling.

Validated by:

• validate_providers

class prompttrail.models.anthropic.AnthropicMessageDict

Bases: TypedDict

content: str

role: Literal['user', 'assistant', 'system']

pydantic model prompttrail.models.anthropic.AnthropicModel

Bases: Model

Model class for Anthropic Claude API.

Fields:

• client (anthropic.Anthropic | None)

• configuration (prompttrail.models.anthropic.AnthropicConfig)

field client: Anthropic | None = None

field configuration: AnthropicConfig [Required]

format_tool(tool: Tool) → Dict[str, Any]

Convert tool to Anthropic format

format_tool_result(result: ToolResult) → Dict[str, Any]

Format result for Anthropic API

list_models() → List[str]

Return a list of available models.

validate_tools(tools: List[Tool]) → None

Validate tools according to Anthropic API requirements

prompttrail.models.google module

pydantic model prompttrail.models.google.GoogleChatExample

Bases: BaseModel

Example for few-shot learning in Google API.

Fields:

• prompt (str)

• response (str)

field prompt: str [Required]

Example prompt.

field response: str [Required]

Example response.

pydantic model prompttrail.models.google.GoogleConfig

Bases: Config

Integration configuration class for Google API.

Manages authentication credentials and model parameters in a centralized way.

Fields:

• api_key (str)

• candidate_count (int | None)

• context (str | None)

• examples (List[prompttrail.models.google.GoogleChatExample] | None)

• max_tokens (int | None)

• model_name (str)

• temperature (float | None)

• top_k (int | None)

• top_p (float | None)

Validators:

field api_key: str [Required]

API key for Google API.

Validated by:

• validate_providers

field candidate_count: int | None = None

Number of candidate responses.

Validated by:

• validate_providers

field context: str | None = None

Optional context for the model.

Validated by:

• validate_providers

field examples: List[GoogleChatExample] | None = None

Optional few-shot learning examples.

Validated by:

• validate_providers

field max_tokens: int | None = 1024

Maximum output tokens.

Validated by:

• validate_providers

field model_name: str = 'models/gemini-1.5-flash'

Model name. Use list_models() to see available models.

Validated by:

• validate_providers

field temperature: float | None = 1.0

Sampling temperature.

Validated by:

• validate_providers

field top_k: int | None = None

Top-k sampling threshold.

Validated by:

• validate_providers

field top_p: float | None = None

Nucleus sampling threshold.

Validated by:

• validate_providers

pydantic model prompttrail.models.google.GoogleModel

Bases: Model

Google API model implementation.

Fields:

• configuration (prompttrail.models.google.GoogleConfig)

field configuration: GoogleConfig [Required]

list_models() → List[str]

Get list of available model names.

Returns:

List of model names

validate_session(session: Session, is_async: bool = False) → None

Validate session for Google API requirements.

Parameters:

• session – Session to validate

• is_async – Whether validation is for async operation

Raises:

ParameterValidationError – If session is invalid

prompttrail.models.openai module

pydantic model prompttrail.models.openai.OpenAIConfig

Bases: Config

Integration configuration class for OpenAI Chat API.

Manages authentication credentials and model parameters in a centralized way.

Fields:

• api_base (str | None)

• api_key (str)

• api_version (str | None)

• max_tokens (int | None)

• model_name (str)

• organization_id (str | None)

• temperature (float | None)

Validators:

field api_base: str | None = None

Base URL for OpenAI API.

Validated by:

• validate_providers

field api_key: str [Required]

API key for OpenAI API.

Validated by:

• validate_providers

field api_version: str | None = None

API version for OpenAI API.

Validated by:

• validate_providers

field max_tokens: int | None = 100

Maximum number of tokens to generate.

Validated by:

• validate_providers

field model_name: str = 'gpt-4o-mini'

Name of the model to use.

Validated by:

• validate_providers

field organization_id: str | None = None

Organization ID for OpenAI API.

Validated by:

• validate_providers

field temperature: float | None = 1.0

Temperature for sampling.

Validated by:

• validate_providers

pydantic model prompttrail.models.openai.OpenAIModel

Bases: Model

Model class for OpenAI Chat API.

Fields:

• configuration (prompttrail.models.openai.OpenAIConfig)

field configuration: OpenAIConfig [Required]

format_tool(tool: Tool) → Dict[str, Any]

Convert tool to OpenAI format.

list_models() → List[str]

Return a list of available models.

prompttrail.models.transformers module

pydantic model prompttrail.models.transformers.TransformersConfig

Bases: Config

Integration configuration class for Transformers models.

Manages authentication credentials and model parameters in a centralized way.

Fields:

• device (str | None)

• max_tokens (int | None)

• model_name (str)

• repetition_penalty (float | None)

• temperature (float | None)

• top_k (int | None)

• top_p (float | None)

Validators:

field device: str | None = None

Device to run model on (e.g. ‘cpu’, ‘cuda’).

Validated by:

• validate_providers

field max_tokens: int | None = 1024

Maximum number of tokens to generate.

Validated by:

• validate_providers

field model_name: str [Required]

Name of the model to use.

Validated by:

• validate_providers

field repetition_penalty: float | None = 1.0

Higher values penalize repeated tokens more strongly.

Validated by:

• validate_providers

field temperature: float | None = 1.0

Sampling temperature between 0 and 1.

Validated by:

• validate_providers

field top_k: int | None = None

Top-k sampling.

Validated by:

• validate_providers

field top_p: float | None = 1.0

Nucleus sampling probability.

Validated by:

• validate_providers

pydantic model prompttrail.models.transformers.TransformersModel

Bases: Model

Model class for running transformer models locally.

Parameters:

• configuration – Model configuration.

• model – Pre-trained transformer model.

• tokenizer – Tokenizer for the model.

Fields:

• model (transformers.models.auto.modeling_auto.AutoModelForCausalLM | None)

• tokenizer (transformers.models.auto.tokenization_auto.AutoTokenizer | None)

field model: AutoModelForCausalLM | None = None

field tokenizer: AutoTokenizer | None = None

__init__(configuration: TransformersConfig, model: AutoModelForCausalLM, tokenizer: AutoTokenizer)

Create a new model by parsing and validating input data from keyword arguments.

Raises [ValidationError][pydantic_core.ValidationError] if the input data cannot be validated to form a valid model.

self is explicitly positional-only to allow self as a field name.

before_send(session: Session | None, is_async: bool) → Session | None

Perform pre-send processing.

validate_session(session: Session, is_async: bool = False) → None

Validate session for transformer models.

Module contents

Module for various LLM models and providers.

This module contains model abstractions for using different LLM providers: - OpenAI GPT models - Anthropic Claude - Google PaLM - Local models via Transformers

Each provider has its own submodule with Configuration, Model and Parameter classes.