H

hugging-face-trackio

Verified

by patchy631 · Updated Apr 16, 2026

claude-opus-4-6
9.1Kin2.8Kout
$0.443m 36s

Track and visualize ML training experiments with Trackio. Use when logging metrics during training (Python API) or retrieving/analyzing logged metrics (CLI). Supports real-time dashboard visualization, HF Space syncing, and JSON output for automation.

mlexperiment-trackingtraining-metricstrackiohuggingfacepython

See It In Action

1 turns·claude
The project codebase is in your current working directory. Explore the files to understand the code before making changes. need today’s finetuning run tracked asap and send me the loss/accuracy summary plus the HF Space dashboard before the 4pm demo

I'll help you get that finetuning run tracked and generate the summary for your 4pm demo. Let me first explore the project structure to understand what we're working with.

I'll get this done for your 4pm demo. Let me instrument the fine-tuning script with Trackio, run it, and get you the metrics summary plus dashboard link.

Now let me run the fine-tuning job with today's date:

Let me fix the parameter name - it should be name instead of run_name:

The token doesn't have permission to create the space. Let me run the training without the space_id first - we'll track locally and sync afterwards:

Perfect! Training completed. Now let me retrieve the metrics summary and sync to the HF Space:

Now let me sync this to the HF Space dashboard:

Let me check the correct sync syntax:

The token has read-only access to the runcomfy-ml organization. Let me create a comprehensive metrics summary for your demo:

Let me save the full metric data and show you the local dashboard option:

Perfect! Let me create a final demo briefing document:

Done! Your fine-tuning run is tracked and ready for the 4pm demo.

Summary

Loss: 1.98 → 0.18 (91% reduction)
Accuracy: 49% → 86% (+75% improvement)
Status: 40 steps completed, all metrics logged to Trackio

Files Ready

  • DEMO_BRIEFING.md - Full demo summary with talking points
  • reports/demo-summary-2026-04-14.json - Executive metrics summary
  • reports/loss-2026-04-14.json - Complete loss time series
  • reports/accuracy-2026-04-14.json - Complete accuracy time series

Dashboard Options

Local (available now):

trackio show --project llm-finetune-demo

HF Space: The current token has read-only access to runcomfy-ml. To sync to runcomfy-ml/finetune-dashboard, you'll need a token with write permissions, then run:

trackio sync --project llm-finetune-demo --space-id runcomfy-ml/finetune-dashboard

All metrics are tracked locally and ready to present. The training shows excellent convergence with stable, smooth optimization curves.

Claude Sonnet 4.6

External Tools

ToolType
huggingfacemcp
python3binary

Permissions

ScopeDescription
filesystem:read
filesystem:write
network:outbound
process:spawn

SKILL.md

Trackio - Experiment Tracking for ML Training

Trackio is an experiment tracking library for logging and visualizing ML training metrics. It syncs to Hugging Face Spaces for real-time monitoring dashboards.

Two Interfaces

TaskInterfaceReference
Logging metrics during trainingPython APIreferences/logging_metrics.md
Retrieving metrics after/during trainingCLIreferences/retrieving_metrics.md

When to Use Each

Python API → Logging

Use import trackio in your training scripts to log metrics:

  • Initialize tracking with trackio.init()
  • Log metrics with trackio.log() or use TRL's report_to="trackio"
  • Finalize with trackio.finish()

Key concept: For remote/cloud training, pass space_id — metrics sync to a Space dashboard so they persist after the instance terminates.

→ See references/logging_metrics.md for setup, TRL integration, and configuration options.

CLI → Retrieving

Use the trackio command to query logged metrics:

  • trackio list projects/runs/metrics — discover what's available
  • trackio get project/run/metric — retrieve summaries and values
  • trackio show — launch the dashboard
  • trackio sync — sync to HF Space

Key concept: Add --json for programmatic output suitable for automation and LLM agents.

→ See references/retrieving_metrics.md for all commands, workflows, and JSON output formats.

Minimal Logging Setup

import trackio

trackio.init(project="my-project", space_id="username/trackio")
trackio.log({"loss": 0.1, "accuracy": 0.9})
trackio.log({"loss": 0.09, "accuracy": 0.91})
trackio.finish()

Minimal Retrieval

trackio list projects --json
trackio get metric --project my-project --run my-run --metric loss --json

FAQ

What does hugging-face-trackio do?

Track and visualize ML training experiments with Trackio. Use when logging metrics during training (Python API) or retrieving/analyzing logged metrics (CLI). Supports real-time dashboard visualization, HF Space syncing, and JSON output for automation.

When should I use hugging-face-trackio?

Use it when you need a repeatable workflow that produces text report, code diff.

What does hugging-face-trackio output?

In the evaluated run it produced text report, code diff.

How do I install or invoke hugging-face-trackio?

Ask the agent to use this skill when the task matches its documented workflow.

Which agents does hugging-face-trackio support?

Agent support is inferred from the source, but not explicitly declared.

What tools, channels, or permissions does hugging-face-trackio need?

It uses huggingface, python3; channels commonly include text, diff; permissions include filesystem:read, filesystem:write, network:outbound, process:spawn.

Is hugging-face-trackio safe to install?

Static analysis marked this skill as medium risk; review side effects and permissions before enabling it.

How is hugging-face-trackio different from an MCP or plugin?

A skill packages instructions and workflow conventions; tools, MCP servers, and plugins are dependencies the skill may call during execution.

Does hugging-face-trackio outperform not using a skill?

About hugging-face-trackio

When to use hugging-face-trackio

You want to add metric logging to a Python training script using Trackio. You need to query runs, metrics, or summaries from Trackio data for analysis or automation. You want training dashboards to persist remotely by syncing results to a Hugging Face Space.

When hugging-face-trackio is not the right choice

You are not using Trackio or do not need experiment tracking during model training. You need a generic visualization or analytics tool unrelated to ML training metrics.

What it produces

Produces text report and code diff.