---
license: other
tags:
  - protein-language-models
  - model-registry
  - bioinformatics
  - protein-design
configs:
- config_name: models
  data_files:
  - split: train
    path: models.parquet
- config_name: aliases
  data_files:
  - split: train
    path: aliases.parquet
- config_name: benchmarks
  data_files:
  - split: train
    path: benchmarks.parquet
---

# BIIE-AI Protein Model Registry

A curated, structured registry of protein language models (PLMs), diffusion generators, structure predictors, inverse folding models, nucleotide language models, and biomedical NLP models.

**Canonical local source:** `registry_src/models.jsonl` (one record per line, edit here)
**Published format:** Parquet (generated by `scripts/build_parquet.py`)
**Schema:** `schema/model_registry.schema.json`
**Selection policy:** `registry_src/selection_policy.yaml`

## Dataset configs

| Config | Description |
|---|---|
| `models` | One row per model — all registry fields |
| `aliases` | Natural-language aliases → `registry_id` mappings |
| `benchmarks` | Task/metric/value evaluation results (populated over time) |

## Key fields

| Field | Description |
|---|---|
| `registry_id` | Unique snake_case identifier |
| `hf_repo_id` | HuggingFace Hub `org/model` path |
| `category` | Functional type (see taxonomy.md) |
| `primary_tasks` | Main use cases |
| `compute_tier` | GPU requirements |
| `commercial_use` | Licensing assessment |
| `selection_tags` | Tags used by the model-selection skill for routing |

## Workflow

```bash
# Edit registry
vim registry_src/models.jsonl

# Validate
python scripts/validate_registry.py

# Build Parquet
python scripts/build_parquet.py

# Sync live HF metadata (downloads, likes, gating)
python scripts/refresh_hf_metadata.py

# Publish to this repo
HF_TOKEN=hf_... python scripts/publish_to_hub.py
```

## Taxonomy reference

See `docs/taxonomy.md` for all enum definitions (`category`, `architecture_type`, `compute_tier`, `commercial_use`, etc.).