README.md

---
license: openmdw-1.0
pipeline_tag: feature-extraction
tags:
- vision-language
- multimodal
- image-text-retrieval
- hyperbolic-embeddings
- clip
- research
- scratch-training
- hier-beta
- argent
---

# Hyper3-CLIP beta

Hyper3-CLIP beta is the hyper³labs ViT-B scratch checkpoint trained with the
hier-beta ARGENT objective.

This repository publishes the raw PyTorch training checkpoint for the completed
500k-step paper-scratch run. It is not the older Hyper3-CLIP v0.5
SentenceTransformers package.

## Artifact

- Checkpoint: `checkpoint_final.pt`
- Config: `config.yaml`
- Training metadata: `metadata.json`
- Run: `hyper3_vitb_clip_uncha_hier_beta_argent_mp5_paper_scratch_8x500k_s31`
- Objective: `uncha` with `uncha_entailment_loss: hier_beta_argent`
- Vision backbone: `vit_base_patch16_224`
- Vision pretrained: `false`
- Text model architecture/tokenizer: `openai/clip-vit-base-patch32`
- Text pretrained: `false`
- Embedding dimension: 512
- Training steps: 500,000
- Global batch size: 768

## Evaluation

The `eval/` directory includes the paper-comparable full benchmark table and the
raw wide summary row used for the current model comparison.

Headline row from the local full eval:

- ImageNet top-1: 46.984%
- COCO I2T/T2I R@10: 84.30 / 73.19
- Flickr I2T/T2I R@10: 97.60 / 91.44
- WordNet hierarchy: TIE 3.1597, LCA 2.0786, Jaccard 0.8179
- PEP AUC/AP: 96.07 / 69.36

The checkpoint is strong on retrieval in the paper-comparable table, but weak on
several flat/fine-grained zero-shot datasets such as Food101, CUB, Flowers102,
Cars, and Aircraft. Treat this release as a research checkpoint, not a polished
production model.

## Loading

This is a raw training checkpoint. Use the hyper³labs `hyper3-clip` codebase and
the included `config.yaml` to instantiate the model, then load
`checkpoint_final.pt`.

```python
import torch

checkpoint = torch.load("checkpoint_final.pt", map_location="cpu", weights_only=False)
state_dict = checkpoint.get("model", checkpoint)
```

## License And Attribution

The model materials in this repository are released under OpenMDW-1.0.
Redistributions should preserve `NOTICE`, `LICENSE`, and the model card when
practical.

Please cite and link to the original hyper³labs model repository when publishing
benchmarks, papers, derivative checkpoints, or public demos based on this model.