Pipeline

Pipeline is the simplest inference API for Hub models across many tasks.

Adapted for this playbook from the 🤗 Transformers documentation by Hugging Face. Official page: https://huggingface.co/docs/transformers/pipeline_tutorial. Images © Hugging Face (documentation-images) unless noted. This is not a substitute for the upstream docs — verify against the current version.

Covers Hugging Face pages: pipeline_tutorial, add_new_pipeline

Executive summary

Use Pipeline when you want preprocessing + model + postprocessing in one call. Override model, pick a task, and accelerate with GPU / half precision when needed.

Minimal example

from transformers import pipeline

pipe = pipeline(task="text-generation", model="openai-community/gpt2")
print(pipe("the secret to baking a really good cake is ", max_new_tokens=30))

Cross-modality examples

ASR

asr = pipeline(task="automatic-speech-recognition", model="openai/whisper-large-v3")
asr("https://huggingface.co/datasets/Narsil/asr_dummy/resolve/main/mlk.flac")

Image classification

clf = pipeline(task="image-classification", model="google/vit-base-patch16-224")
clf("https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/pipeline-cat-chonk.jpeg")

Source: Hugging Face documentation images (pipeline-cat-chonk.jpeg).

Useful parameters

Parameter	Role
task	Selects the task-specific pipeline class
model	Checkpoint id or local path
device / Accelerate	GPU / MPS placement
torch_dtype	float16 / bfloat16 for memory and speed
batch_size	Throughput for lists of inputs

When Pipeline is not enough

Move to AutoModel + generate for custom logits processors, advanced caching, or tight latency loops — see Text generation.

Key takeaways

• Pipeline is the right default for demos, batch jobs and many production task APIs.
• Always set generation limits (max_new_tokens) for text tasks.

What to do next

Pipeline apps and web servers or Text generation.