akhilshaik07/everydayness-agent-traces

Everydayness — Anonymized Agent Traces

Anonymized execution traces from the Everydayness grief companion app, built for the Build Small Hackathon (Gradio × Hugging Face, June 2026).

What this dataset contains

Each line in anonymized_traces.jsonl is one pipeline execution trace.

What IS logged:

• session_id — UUID v4, no user info attached
• pipeline — which pipeline ran (pain_filter | audio | media | session_summary)
• model — model name and quantization
• batch_size — number of messages per chunk
• classification_distribution — label ratios (no message content)
• inference_latency_ms — how long the Modal call took
• gpu — GPU type used
• prompt_template_summary — template description with {placeholders}

What is NEVER logged:

• Message content
• Sender names
• Chat timestamps
• Any personally identifiable information
• File contents or transcripts

Why this exists

This dataset earns the 📡 Sharing is Caring badge for the Build Small Hackathon. By sharing anonymized pipeline traces publicly, we help the community understand how small models (Qwen3-14B, Whisper) perform on real grief-companion classification tasks.

App

• HF Space: build-small-hackathon/everydayness
• Track: Backyard AI
• Model: OpenPipe/Qwen3-14B-Instruct (AWQ 4-bit, xgrammar, Modal A100)