Model Discovery Dashboard

HuggingFace model discovery and validation pipeline

Task Queue Status

-
Pending Tasks
-
Processing
-
Failed
Recent Discovery Tasks
Last updated: -
Task ID Use-Case Status Priority Created Error
Loading...

Model Candidates

-
Validated
-
Submitted to Arena
-
Rejected
Model Candidates
Model Validation Status Metadata Discovered Issues
Loading...

Use-Case Catalog Observability

-
Alias Resolutions
-
Invalid Alias Attempts
-
Curated Injections
Invalid Alias Details

Shows alias spellings that fell back to the default use case.

  • No invalid aliases recorded yet.
Curated Recommendation Breakdown

Counts of catalog-curated candidates injected before live search results.

  • No curated recommendations injected yet.
How it works:
  1. Tasks are created with weighted priorities based on use-case demand
  2. HuggingFace API is queried (rate-limited to 1 call/5s)
  3. Candidates are validated against 5 criteria (size, architecture, license, activity, tags)
  4. Valid candidates are batch-submitted to Model Arena for evaluation
  5. Results are auto-appended to feature roadmap

Discovery Agent Workflow

Autonomous Discovery Agent

The Discovery Agent runs daily at 6 AM UTC, searching HuggingFace for model candidates, conducting deep research with GPT-Researcher, and auto-submitting ALL validated models to Arena.

-
Runs Today
-
Models Discovered
-
Submitted to Arena
-
Last Run
Recent Agent Runs
Execution ID Use Cases Status Discovered Submitted Started
No recent agent runs
KeikenV Model Discovery • Automated Model Scouting