kshitijthakkar/smoltrace-leaderboard

Tiny Agents. Total Visibility.

---

SMOLTRACE Leaderboard

This dataset contains aggregated evaluation metrics for comparing model performance across SMOLTRACE benchmark runs.

Dataset Information

Field	Value
Owner	kshitijthakkar
Updated	2025-12-01 13:59:33 UTC
Purpose	Model comparison and ranking

Schema

Identification

Column	Type	Description
run_id	string	Unique run identifier
model	string	Model name/identifier
agent_type	string	Agent type ("tool", "code", "both")
provider	string	Model provider (litellm, openai, etc.)
timestamp	string	Evaluation timestamp
submitted_by	string	HuggingFace username

Dataset References

Column	Type	Description
results_dataset	string	Link to results dataset
traces_dataset	string	Link to traces dataset
metrics_dataset	string	Link to metrics dataset
dataset_used	string	Source benchmark dataset

Performance Metrics

Column	Type	Description
total_tests	int	Number of test cases
successful_tests	int	Passed tests
failed_tests	int	Failed tests
success_rate	float	Success percentage (0-100)
avg_steps	float	Average agent steps per test
avg_duration_ms	float	Average execution time (ms)
total_duration_ms	float	Total evaluation time (ms)
total_tokens	int	Total tokens consumed
avg_tokens_per_test	int	Average tokens per test
total_cost_usd	float	Total API cost (USD)
avg_cost_per_test_usd	float	Average cost per test (USD)

Environmental Impact

Column	Type	Description
co2_emissions_g	float	Total CO2 emissions (gCO2e)
power_cost_total_usd	float	Total power cost (USD)

GPU Metrics (if available)

Column	Type	Description
gpu_utilization_avg	float	Average GPU utilization (%)
gpu_utilization_max	float	Peak GPU utilization (%)
gpu_memory_avg_mib	float	Average GPU memory (MiB)
gpu_memory_max_mib	float	Peak GPU memory (MiB)
gpu_temperature_avg	float	Average GPU temperature (°C)
gpu_temperature_max	float	Peak GPU temperature (°C)
gpu_power_avg_w	float	Average GPU power (W)

Usage

from datasets import load_dataset
import pandas as pd

# Load leaderboard
ds = load_dataset("kshitijthakkar/smoltrace-leaderboard")
df = pd.DataFrame(ds['train'])

# Rank by success rate
df_ranked = df.sort_values('success_rate', ascending=False)
print(df_ranked[['model', 'success_rate', 'avg_duration_ms', 'total_cost_usd']])

# Compare models
top_models = df_ranked.head(10)
print("Top 10 Models by Success Rate:")
for i, row in top_models.iterrows():
    print(f"  {row['model']}: {row['success_rate']:.1f}%")

Contributing Results

Run your own evaluations to add to this leaderboard:

pip install smoltrace
smoltrace-eval --model your-model --provider litellm

---

About SMOLTRACE

SMOLTRACE is a comprehensive benchmarking and evaluation framework for Smolagents - HuggingFace's lightweight agent library.

Key Features

• Automated agent evaluation with customizable test cases
• OpenTelemetry-based tracing for detailed execution insights
• GPU metrics collection (utilization, memory, temperature, power)
• CO2 emissions and power cost tracking
• Leaderboard aggregation and comparison

Quick Links

• GitHub Repository
• PyPI Package
• Documentation
• Report Issues

Installation

pip install smoltrace

Citation

If you use SMOLTRACE in your research, please cite:

@software{smoltrace,
  title = {SMOLTRACE: Benchmarking Framework for Smolagents},
  author = {Thakkar, Kshitij},
  url = {https://github.com/Mandark-droid/SMOLTRACE},
  year = {2025}
}

---

_{Generated by SMOLTRACE}