llm4s-core/org.llm4s/org.llm4s.model/RequestTransformer

RequestTransformer

org.llm4s.model.RequestTransformer

See theRequestTransformer companion object

Handles model-specific parameter validation and transformation.

Uses ModelCapabilities from ModelRegistry to apply constraints based on what each model supports. This mirrors LiteLLM's approach to handling model-specific quirks (e.g., O-series temperature restrictions).

Example usage:

 val transformer = RequestTransformer.default
 val result = transformer.transformOptions("o1", options, dropUnsupported = true)
 result match {
   case Right(transformed) => // use transformed options
   case Left(error) => // handle validation error
 }

Attributes

Companion: object
Graph
Supertypes: class Object

trait Matchable

class Any
Known subtypes: class DefaultRequestTransformer

Members list

Value members

Abstract methods

Get the set of parameters that are not supported by this model.

Value parameters

modelId: The model identifier

Attributes

Returns: Set of disallowed parameter names, empty if all are allowed

Check if streaming needs to be faked for this model.

Some models (like O1) don't support native streaming and require the client to simulate streaming by returning the full response as a single chunk.

Value parameters

modelId: The model identifier

Attributes

Returns: true if the model requires fake streaming

Check if this model requires max_completion_tokens instead of max_tokens.

Some models (like O-series reasoning models) require the use of max_completion_tokens parameter instead of the standard max_tokens.

Value parameters

modelId: The model identifier

Attributes

Returns: true if the model requires max_completion_tokens

Transform messages for model-specific requirements.

For example, O-series models that don't support system messages will have their system messages converted to user messages with a "[System]:" prefix.

Value parameters

messages: The messages to transform
modelId: The model identifier

Attributes

Returns: Transformed messages

Validate and transform completion options for a specific model.

Checks model capabilities and either drops unsupported parameters or returns validation errors, depending on the dropUnsupported flag.

Value parameters

dropUnsupported: If true, silently drop/adjust unsupported params; if false, return error
modelId: The model identifier (e.g., "o1", "gpt-4o", "claude-3-7-sonnet")
options: The completion options to transform

Attributes

Returns: Transformed options or validation error

In this article

Generated with