Implementation:Cohere ai Cohere python RerankerDataMetrics Model

Knowledge Sources	Cohere Python SDK
Domains	SDK, Fine-tuning
Last Updated	2026-02-15 14:00 GMT

Overview

RerankerDataMetrics is a Pydantic model that contains metrics about the training and evaluation datasets used for reranker fine-tuning.

Description

The RerankerDataMetrics class extends UncheckedBaseModel and provides six optional integer fields describing the composition of reranker fine-tuning datasets. The training metrics include the number of training queries (num_train_queries), the total count of relevant passages across training examples (num_train_relevant_passages), and the total count of hard negatives across training examples (num_train_hard_negatives). The evaluation metrics mirror the same structure: num_eval_queries, num_eval_relevant_passages, and num_eval_hard_negatives. All fields are optional and default to None.

Usage

Use RerankerDataMetrics when inspecting the dataset composition of a reranker fine-tuning job. This model provides insight into the size and balance of training and evaluation data, which is useful for monitoring fine-tuning quality, diagnosing data imbalances, or validating that the training dataset meets minimum requirements before initiating a fine-tuning run.

Code Reference

Source Location

• Repository: Cohere Python SDK
• File: src/cohere/types/reranker_data_metrics.py

Signature

class RerankerDataMetrics(UncheckedBaseModel):
    num_train_queries: typing.Optional[int] = pydantic.Field(default=None)
    num_train_relevant_passages: typing.Optional[int] = pydantic.Field(default=None)
    num_train_hard_negatives: typing.Optional[int] = pydantic.Field(default=None)
    num_eval_queries: typing.Optional[int] = pydantic.Field(default=None)
    num_eval_relevant_passages: typing.Optional[int] = pydantic.Field(default=None)
    num_eval_hard_negatives: typing.Optional[int] = pydantic.Field(default=None)

Import

from cohere.types import RerankerDataMetrics

I/O Contract

Fields

Field	Type	Required	Default	Description
num_train_queries	Optional[int]	No	None	The number of training queries.
num_train_relevant_passages	Optional[int]	No	None	The sum of all relevant passages across valid training examples.
num_train_hard_negatives	Optional[int]	No	None	The sum of all hard negatives across valid training examples.
num_eval_queries	Optional[int]	No	None	The number of evaluation queries.
num_eval_relevant_passages	Optional[int]	No	None	The sum of all relevant passages across valid evaluation examples.
num_eval_hard_negatives	Optional[int]	No	None	The sum of all hard negatives across valid evaluation examples.

Usage Examples

from cohere.types import RerankerDataMetrics

# RerankerDataMetrics is typically returned as part of a fine-tuning job response

# Example: inspecting metrics from a fine-tuning job
metrics = RerankerDataMetrics(
    num_train_queries=1500,
    num_train_relevant_passages=4200,
    num_train_hard_negatives=3000,
    num_eval_queries=300,
    num_eval_relevant_passages=850,
    num_eval_hard_negatives=600,
)

# Display training data summary
print(f"Training queries: {metrics.num_train_queries}")
print(f"Training relevant passages: {metrics.num_train_relevant_passages}")
print(f"Training hard negatives: {metrics.num_train_hard_negatives}")

# Display evaluation data summary
print(f"Evaluation queries: {metrics.num_eval_queries}")
print(f"Evaluation relevant passages: {metrics.num_eval_relevant_passages}")
print(f"Evaluation hard negatives: {metrics.num_eval_hard_negatives}")

# Calculate average passages per query
if metrics.num_train_queries and metrics.num_train_relevant_passages:
    avg_passages = metrics.num_train_relevant_passages / metrics.num_train_queries
    print(f"Avg relevant passages per training query: {avg_passages:.1f}")

Related Pages

• Environment:Cohere_ai_Cohere_python_Python_SDK_Runtime