Validation Error: LLM returns float instead of List[Object] in nested Pydantic models

[psharma28]

Issue Description

When using Instructor with Anthropic/Claude model, I'm encountering validation errors where the LLM returns single float values instead of expected List[Object] structures.

Code Example

# Models
class RecipeStep(BaseModel):
    step_number: str = Field(..., description="Step number in the cooking process")
    duration: str = Field(..., description="Time required for this step")
    temperature: float = Field(..., description="Required temperature")

class CookingMethod(BaseModel):
    method_name: str = Field(..., description="Name of cooking method")
    steps: List[RecipeStep] = Field(..., description="Details of cooking steps")

class DishDetails(BaseModel):
    dish_name: str = Field(..., description="Name of the dish")
    cooking_approaches: List[CookingMethod]

class RecipeResponse(BaseModel):
    chef_explanation: str
    dish_details: List[DishDetails]

# Usage
response = anthropic_client.create(
    model="claude-3-sonnet-20240229",
    messages=[{"role": "user", "content": content}],
    response_model=RecipeResponse
)

Error Message

instructor.exceptions.InstructorRetryException: 15 validation errors for RecipeResponse
dish_details.0.cooking_approaches.0.steps
Input should be a valid array [type=list_type, input_value=350.5, input_type=float]

Expected Behavior

The LLM should return a properly structured object with steps as a List[RecipeStep] containing step objects with number, duration, and temperature.

Actual Behavior

Instead, the LLM is returning single float values where List[RecipeStep] is expected.

Environment

instructor version: 1.7.2
pydantic version: 2.10.6
Python version: 3.12
Model: Claude 3.5 V2 through AnthropicBedrock

Question

1. Is this a known limitation with the current implementation?
2. Are there recommended approaches to enforce the expected list structure?

Any guidance would be appreciated!