Spaces:

---

FlameF0X

/

Tiny-T2I-Leaderboard

like 4

Running

>>──────────────────────────────────────────────────────────────────────────────<<
To add your model, open a PR adding a new object to the models array below.
Copy this template and fill in your values:

{
   name:        "YourModel-1B",        // Required — model name
   org:         "your-org",            // Required — unique org key (lowercase kebab)
   params:      "1.0B",                // Required — e.g. "50M", "1.2B", "800M" it has to be the total parameter count
   fid:         12.34,                 // Required — FID score (lower is better)
   clipScore:   0.321,                 // Required — CLIP Score (0–1, higher better)
   resolution:  "512x512",             // Optional — training/generation resolution
   releaseDate: "2026-07-13",          // Optional — ISO date string YYYY-MM-DD
   links: {
     card: "https://huggingface.co/your-org/your-model",   // Optional — model card
     demo: "https://huggingface.co/spaces/..."             // Optional — interactive demo
   }
},

Then add entries in the three maps below so the UI colors your org correctly:

orgNameMap["your-org"] = "Your Org";          // Display name
colorMap["your-org"]   = "#ff6600";           // Brand color (hex)

>>──────────────────────────────────────────────────────────────────────────────<<

For harness use:

Diffusers Bench — Hugging Face's evaluation suite using the Diffusers library
T2I-CompBench — Compositional text-to-image benchmark
MJHQ — Midjourney-style Human Quality evaluation
COCO Captions — Using COCO's ground-truth captions with FID (Fréchet Inception Distance) and CLIP Score as metrics