# Huffing Face 😤

> A very fast mirror of permissively-licensed open-weights models.
> Same files, same revision hashes, same checksums as huggingface.co — just closer.
> Not affiliated with Hugging Face. We just huff models.

## Use

```sh
export HF_ENDPOINT=https://huffingface.co
# then transformers / huggingface_hub / hf work unchanged
```

Files we have are served locally at high speed; anything else 302s to
huggingface.co and gets cached for the next requester (if its license
permits redistribution — gated/restricted repos are never cached).

## API

- `GET  https://huffingface.co/api/v1/models` — catalog (JSON)
- `GET  https://huffingface.co/api/v1/models/{org}/{repo}` — files + cache state
- `POST https://huffingface.co/api/v1/requests` `{"repo": "org/name"}` — ask us to mirror a model
- `GET  https://huffingface.co/{org}/{repo}/resolve/{revision}/{file}` — HF-compatible download

## Mirrored models

- [BAAI/bge-base-en-v1.5](https://huffingface.co/BAAI/bge-base-en-v1.5) — mit, 1.3 GB, fully huffed
- [BAAI/bge-m3](https://huffingface.co/BAAI/bge-m3) — mit, 4.6 GB, fully huffed
- [BAAI/bge-reranker-v2-m3](https://huffingface.co/BAAI/bge-reranker-v2-m3) — apache-2.0, 2.3 GB, fully huffed
- [BAAI/bge-small-en-v1.5](https://huffingface.co/BAAI/bge-small-en-v1.5) — mit, 0.4 GB, fully huffed
- [Qwen/Qwen2.5-0.5B-Instruct](https://huffingface.co/Qwen/Qwen2.5-0.5B-Instruct) — apache-2.0, 1.0 GB, fully huffed
- [Qwen/Qwen2.5-1.5B-Instruct](https://huffingface.co/Qwen/Qwen2.5-1.5B-Instruct) — apache-2.0, 3.1 GB, fully huffed
- [Qwen/Qwen2.5-7B-Instruct](https://huffingface.co/Qwen/Qwen2.5-7B-Instruct) — apache-2.0, 15.2 GB, fully huffed
- [Qwen/Qwen3-0.6B](https://huffingface.co/Qwen/Qwen3-0.6B) — apache-2.0, 1.5 GB, fully huffed
- [Qwen/Qwen3-32B](https://huffingface.co/Qwen/Qwen3-32B) — apache-2.0, 65.5 GB, fully huffed
- [Qwen/Qwen3-4B](https://huffingface.co/Qwen/Qwen3-4B) — apache-2.0, 8.1 GB, fully huffed
- [Qwen/Qwen3-8B](https://huffingface.co/Qwen/Qwen3-8B) — apache-2.0, 16.4 GB, fully huffed
- [Qwen/Qwen3-Embedding-0.6B](https://huffingface.co/Qwen/Qwen3-Embedding-0.6B) — apache-2.0, 1.2 GB, fully huffed
- [deepseek-ai/DeepSeek-R1-Distill-Llama-8B](https://huffingface.co/deepseek-ai/DeepSeek-R1-Distill-Llama-8B) — mit, 16.1 GB, fully huffed
- [deepseek-ai/DeepSeek-R1-Distill-Qwen-7B](https://huffingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-7B) — mit, 15.2 GB, fully huffed
- [empero-ai/Qwythos-9B-Claude-Mythos-5-1M-GGUF](https://huffingface.co/empero-ai/Qwythos-9B-Claude-Mythos-5-1M-GGUF) — apache-2.0, 96.2 GB, huffing
- [google-bert/bert-base-uncased](https://huffingface.co/google-bert/bert-base-uncased) — apache-2.0, 3.5 GB, fully huffed
- [mistralai/Mistral-7B-Instruct-v0.3](https://huffingface.co/mistralai/Mistral-7B-Instruct-v0.3) — apache-2.0, 29.0 GB, fully huffed
- [nomic-ai/nomic-embed-text-v1.5](https://huffingface.co/nomic-ai/nomic-embed-text-v1.5) — apache-2.0, 2.2 GB, fully huffed
- [nvidia/GLM-5.2-NVFP4](https://huffingface.co/nvidia/GLM-5.2-NVFP4) — mit, 464.9 GB, fully huffed
- [openai/gpt-oss-120b](https://huffingface.co/openai/gpt-oss-120b) — apache-2.0, 195.8 GB, huffing
- [openai/gpt-oss-20b](https://huffingface.co/openai/gpt-oss-20b) — apache-2.0, 41.3 GB, huffing
- [sentence-transformers/all-MiniLM-L6-v2](https://huffingface.co/sentence-transformers/all-MiniLM-L6-v2) — apache-2.0, 1.0 GB, fully huffed
- [sentence-transformers/all-mpnet-base-v2](https://huffingface.co/sentence-transformers/all-mpnet-base-v2) — apache-2.0, 3.8 GB, fully huffed
- [zai-org/GLM-5.2](https://huffingface.co/zai-org/GLM-5.2) — mit, 1506.7 GB, huffing
- [zai-org/GLM-5.2-FP8](https://huffingface.co/zai-org/GLM-5.2-FP8) — mit, 755.7 GB, huffing

## Takedown

Rights holder? abuse@huffingface.co — actioned promptly.