Output formats

Every command renders through one formatter, so the same flags work everywhere. Pick a format with -o, or let hf choose: a table when it is writing to a terminal, JSONL when it is writing to a pipe.

Formats

hf models --author google -o table      # a bordered grid for reading
hf models --author google -o list       # one section per record, good for detail
hf models --author google -o markdown   # a GitHub table to paste somewhere
hf models --author google -o jsonl      # one JSON object per line
hf models --author google -o json       # a single JSON array
hf models --author google -o csv        # spreadsheet friendly
hf models --author google -o tsv        # tab separated
hf models --author google -o url        # the hub URL of each record
hf cat google-bert/bert-base-uncased -o raw   # the bytes, unformatted

Format	Best for
table	Reading a list on a terminal
list	Reading one record, where a grid would be too wide
markdown	Pasting into a doc, an issue, or a README
jsonl	Piping into another tool, one object at a time
json	Loading a whole result as an array
csv / tsv	Spreadsheets and quick column math
url	Feeding hub URLs into other commands
raw	Response bodies and file contents

Columns are a summary, JSON is everything

A record models everything the hub said about a thing, which for a model is around sixty fields and for an org page is more. That does not fit across a screen, so each record type marks the handful of fields worth a column and hides the rest:

hf models --author google -o csv | head -1
id,gated,likes,downloads,modified,task,library

Hidden means hidden from table, list, markdown, csv, and tsv only. The JSON formats always carry the whole record, so a field you cannot see in the table is still one field away in a pipe:

hf model google-bert/bert-base-uncased -o json | jq '.[0].safetensors'

Narrowing and widening columns

--fields names exactly the columns you want, in the order you want them, and a name can be one of the hidden fields:

hf models --author google --fields id,downloads
hf models --author google --fields id,trendingScore,createdAt

The names are the JSON keys. A name the record does not have comes back as an empty column, so a typo shows up as a blank rather than shifting the row.

--no-header drops the header row, which helps when a downstream tool expects bare rows.

Templating rows

For full control over each line, apply a Go text/template. The template runs against the record's JSON, so the names are the JSON keys, lowercase, the same ones jq would use:

hf models --author google --template '{{.id}} has {{.downloads}} downloads'
hf datasets -n 5 --template '{{.id}}: {{.likes}} likes'

A name the record does not carry prints <no value> rather than failing, which is usually a JSON key spelled as a Go field name.

Why auto-detection helps

Because the default adapts to the destination, the same command reads well by hand and parses cleanly in a pipe:

hf models --author google       # a table, because this is a terminal
hf models --author google | jq  # JSONL, because this is a pipe

You only reach for -o when you want something other than that default.