Skip to main content

Weave Traces Analysis Report

Detailed analysis of Weave traces from the mcp-tests project, focusing on operation distribution and token usage.
Created on March 21|Last edited on March 21

Weave Traces Analysis

Overview

This report provides insights into the Weave traces captured in the wandb-applied-ai-team/mcp-tests project on March 12, 2025.

Key Metrics

- **Total Traces**: 10 - **Total Tokens**: 70,734 - **Input Tokens**: 69,298 (98%) - **Output Tokens**: 1,436 (2%) - **Time Range**: March 12, 2025, 14:19:36 - 14:19:44 UTC

Operation Distribution

The traces involve multiple types of operations, with the following distribution: - OpenAI Chat Completions: 2 traces - Async OpenAI LLM Model: 2 traces - WandBot Correctness Evaluator: 2 traces - WandBot Correctness Scorer: 2 traces - Scorer Summarize: 1 trace - Evaluation Summarize: 1 trace

Observations

1. The majority of computational resources are dedicated to input processing. 2. The traces appear to be part of an evaluation process for a support chatbot. 3. The operations are tightly clustered in a very short time frame (less than 10 seconds).

Run set
38



Run set
38