Huffing Face

Models. Inhaled.

A very fast mirror of the open-weights models you were going to download anyway. Same files, same checksums, same tools — just closer.

$ export HF_ENDPOINT=https://huffingface.co

That's it. That's the integration. transformers, huggingface_hub, hf — they all just work, only faster. Anything we haven't huffed yet transparently comes from the source.

42 models
9.35 TB huffed
411 huffs served
40 Gbps of exhale
ModelLicenseSizeHuffsStatus
OpenMOSS-Team/MOSS-Transcribe-Diarizeapache-2.01.8 GB42fully huffed
BAAI/bge-m3mit4.6 GB24fully huffed
BAAI/bge-reranker-v2-m3apache-2.02.3 GB24fully huffed
Qwen/Qwen3-32Bapache-2.065.5 GB20fully huffed
Qwen/Qwen3-8Bapache-2.016.4 GB20fully huffed
Qwen/Qwen3-4Bapache-2.08.1 GB20fully huffed
Qwen/Qwen3-0.6Bapache-2.01.5 GB20fully huffed
thinkingmachines/Inklingapache-2.01.90 TB19fully huffed
sentence-transformers/all-mpnet-base-v2apache-2.03.8 GB19fully huffed
sentence-transformers/all-MiniLM-L6-v2apache-2.0976.9 MB17fully huffed
open-gigaai/Giga-World-1apache-2.0178.9 GB12fully huffed
mistralai/Mistral-7B-Instruct-v0.3apache-2.029.0 GB11fully huffed
zai-org/GLM-5.2mit1.51 TB10fully huffed
zai-org/GLM-5.2-FP8mit755.7 GB10fully huffed
nvidia/GLM-5.2-NVFP4mit464.9 GB10fully huffed
nomic-ai/nomic-embed-text-v1.5apache-2.02.2 GB10fully huffed
openai/gpt-oss-120bapache-2.0195.8 GB9fully huffed
openai/gpt-oss-20bapache-2.041.3 GB9fully huffed
Qwen/Qwen2.5-7B-Instructapache-2.015.2 GB9fully huffed
deepseek-ai/DeepSeek-R1-Distill-Llama-8Bmit16.1 GB8fully huffed
Qwen/Qwen2.5-1.5B-Instructapache-2.03.1 GB8fully huffed
Qwen/Qwen2.5-0.5B-Instructapache-2.0999.6 MB8fully huffed
BAAI/bge-small-en-v1.5mit401.1 MB8fully huffed
empero-ai/Qwythos-9B-Claude-Mythos-5-1M-GGUFapache-2.096.2 GB7fully huffed
deepseek-ai/DeepSeek-R1-Distill-Qwen-7Bmit15.2 GB6fully huffed
google-bert/bert-base-uncasedapache-2.03.5 GB6fully huffed
BAAI/bge-base-en-v1.5mit1.3 GB6fully huffed
Qwen/Qwen3.6-35B-A3Bapache-2.071.9 GB5fully huffed
Qwen/Qwen3.6-27Bapache-2.055.6 GB5fully huffed
moonshotai/Kimi-K3kimi-k31.56 TB4fully huffed
deepseek-ai/DeepSeek-V4-Flashmit159.6 GB4fully huffed
Qwen/Qwen3-Embedding-0.6Bapache-2.01.2 GB4fully huffed
tencent/Hy3apache-2.0597.6 GB3fully huffed
deepreinforce-ai/Ornith-1.0-35Bmit70.3 GB3fully huffed
hexgrad/Kokoro-82Mapache-2.0363.3 MB3fully huffed
baidu/Unlimited-OCRmit6.8 GB2fully huffed
openbmb/MiniCPM5-1Bapache-2.02.2 GB2fully huffed
deepseek-ai/DeepSeek-V4-Promit864.7 GB1fully huffed
moonshotai/Kimi-K2.7-Codemodified-mit595.2 GB1fully huffed
microsoft/TRELLIS.2-4Bmit16.2 GB1fully huffed
lightonai/LightOnOCR-2-1Bapache-2.02.0 GB1fully huffed
Qwen/Qwen3-TTS-12Hz-1.7B-CustomVoiceapache-2.04.5 GB0fully huffed

42 of 42 fully local.

Request a model

If its license lets us redistribute it, your request primes the mirror for everyone.

Huffing Face — fast, HF-compatible open-weights model mirror