newberry-metrics 0.0.30

A model evaluation tool

Newberry Metrics

A Python package for tracking and estimating AI model token costs and usage metrics.

Latest Version: 0.0.10

Features

Cost Tracking and Estimation

• Model cost calculation per million tokens
• Prompt cost estimation
• Session-based cost tracking
• Support for multiple AI models:
  • Claude 3.7 Sonnet
  • Nova Micro

Installation

pip install newberry_metrics

Usage Examples

Calculate Model Cost

from newberry_metrics import model_cost

# Get cost per million tokens for Nova Micro
nova_cost = model_cost("nova micro")
print(f"Nova Micro cost per million tokens: ${nova_cost}")

# Get cost for Claude 3.7 Sonnet
claude_cost = model_cost("claude 3.7 sonnet")
print(f"Claude 3.7 Sonnet cost per million tokens: ${claude_cost}")

Estimate Prompt Cost

from newberry_metrics import prompt_cost

prompt = "What is the weather in San Francisco?"
cost = prompt_cost(prompt, model="nova micro")
print(f"Estimated prompt cost: ${cost}")

Track Session Costs

from newberry_metrics import session_cost

# Track costs across multiple prompts in a session
session_id = "session_1"
cost1 = session_cost(session_id, "First prompt", model="nova micro")
cost2 = session_cost(session_id, "Second prompt", model="nova micro")
print(f"Total session cost: ${cost2}")  # Shows cumulative cost

Technical Details

Token Estimation

• Uses a simple 4 characters per token estimation rule
• Provides conservative estimates for cost calculation
• Supports both input and output token cost calculation

Recent Updates (v0.0.10)

New Features

• Implemented basic token cost estimation
• Added support for Claude 3.7 Sonnet and Nova Micro models
• Introduced session-based cost tracking
• Added utility functions for cost estimation

Technical Improvements

• Simple token estimation algorithm
• Session cost persistence during runtime
• Model pricing configuration system

Requirements

• Python >= 3.10

Contact & Support

• Developer: Satya-Holbox, Harshika-Holbox
• Email: [e-mail]
• GitHub: SatyaTheG

License

This project is licensed under the MIT License.

---

Note: This package is actively maintained and regularly updated with new features and model support.