A to Z
Glossary
Every concept in this sheet, plus the smaller terms it uses. Select a concept to open its card.
A
- Auto classes
- Auto classes read a repo's config and build the right architecture.
C
- Chat templates
- Each chat model expects messages formatted with its own special tokens.
D
- dtype and device_map
dtype="auto"loads weights in the precision they were saved in, oftenbfloat16, instead offloat32.
G
- generate
model.generateproduces tokens one at a time.
H
- Hub repositories
- Every model, dataset and Space on the Hub is a Git-based repository with an id of the form
owner/name, such asQwen/Qwen3-8B. - huggingface_hub in Python
- The
huggingface_hublibrary downloads single files or whole repos, creates repos, uploads folders and searches the Hub from Python.
L
- Licences and gated models
- Each repo declares a licence, such as Apache 2.0, MIT or a custom model licence with usage conditions.
M
- Model cards
- The README of a model repo: what the model is for, how it was trained, its evaluation results, limitations and intended use, plus YAML metadata such as licence, tags and base model.
- Model outputs and heads
- Calling a model returns an output object:
logitsfor models with a task head,last_hidden_statefor a bareAutoModel, andlosswhen you passlabels.
P
- pipeline
pipeline(task, model=...)bundles preprocessing, the model and postprocessing for a task such as text generation, classification, summarisation or speech recognition.
S
- safetensors
- The default weight format on the Hub: a simple file of tensors with a JSON header, which loads fast and cannot run code.
- Spaces
- Spaces host small apps, usually built with Gradio or as a Docker container, directly from a Hub repo, on free CPU or paid GPU hardware.
T
- The hf CLI and login
- The
hfcommand line tool (the successor tohuggingface-cli) logs in, downloads and uploads files. - The local cache
- Downloads are stored once in a shared cache (by default under
~/.cache/huggingface) and reused by every library.