What huggingface_hub 2.0 changes, and why transformers keeps you on 1.x for now
1 month ago
Hugging Face released huggingface_hub 2.0 on September 24, 2026: a new HTTP library underneath (httpx2) and every feature deprecated in 1.x removed. Who has to change code, why transformers, datasets and diffusers users stay on 1.x for now, and where the official migration guide is. https://github.com/huggingface/huggingface_hub/releases/tag/v2.0.0 https://huggingface.co/docs/huggingface_hub/main/en/concepts/migration_v2 https://github.com/huggingface/huggingface_hub/issues/4802
Ask about this presentation
Chapters
Show transcriptHide transcript
Hugging Face released version 2.0 of huggingface_hub on September 24th. It's the Python library that downloads and publishes models and datasets on the Hub, and transformers, datasets and diffusers all depend on it. It's a major version, so some code written for 1.x will stop working.
Two things changed. The HTTP library it uses to talk to the Hub moves from httpx to its successor, httpx2. And everything Hugging Face marked as deprecated during the 1.x series is gone. The most visible one is the old huggingface-cli command. The command line is just hf now.
The reason is upstream. The httpx project is no longer actively maintained. Swapping the library underneath breaks code that depends on it, so Hugging Face shipped the change as a major version.
So who has to change anything? Mostly people who write code against huggingface_hub itself, especially code that handles its network errors or sets up its own HTTP client. Scripts that still use a deprecated feature, including the old command line, need updating too. And certificates now come from your operating system's trust store by default. A custom certificate bundle set the usual way still works.
If you mostly use Hugging Face through its other libraries, this doesn't reach you yet. As of September 28th, the newest releases of transformers, datasets, diffusers and sentence transformers all require huggingface_hub below 2.0, so installing them keeps you on 1.x. Hugging Face's plan is for those libraries to accept both versions. transformers has merged that change but hasn't released it yet, and it's still under review for datasets and diffusers.
For the exact changes, go to Hugging Face's own migration guide. It's in the huggingface_hub docs, under Migrating to v2.0, and it lists every removed feature with its replacement. The full release notes are on GitHub. That's huggingface_hub 2.0: a new HTTP library, the deprecated features gone, and most of the ecosystem on 1.x for now.
Up next
3:39Transformers runs Mistral Large 4 and trains LoRA on FP8 checkpoints
@hugging-face
1 week ago
2:43huggingface_hub v2.4.0 copies files into Buckets and flags outdated CLI extensions
@hugging-face
1 week ago
3:05huggingface_hub v2.3.0 lists an 800,000-file dataset in 20 seconds, down from 13 minutes
@hugging-face
3 weeks ago
2:27Transformers adds Reflection AI’s Beam and loads models on PyTorch 2.7 again
@hugging-face
3 weeks ago
3:46Transformers v5.20.0 runs Meta's Muse Spark Open, and here is the hardware it needs
@hugging-face
1 month ago
3:18huggingface_hub 2.2.0: hf jobs stats prints a snapshot, and parallel downloads always use hf_xet
@hugging-face
1 month ago
3:12Transformers v5.19.0: EmbeddingGemma 2 and token-dispatch expert parallelism
@hugging-face
1 month ago
2:53huggingface_hub's HfFileSystem now downloads a 2 GB bucket file in seconds
@hugging-face
1 month ago
3:12Transformers v5.18.0: streaming speaker diarization, GGUF mixture-of-experts, FSDP2 with expert parallelism
@hugging-face
1 month ago





