Add sort and limit parameters in hf cache ls

docs/source/en/guides/cli.md

@@ -614,7 +614,7 @@ model/microsoft/UserLM-8b be8f2069189bdf443e554c24e488ff3ff6952691    32.1G 4 da
 Found 1 repo(s) for a total of 1 revision(s) and 32.1G on disk.
 ```
 
-The command supports several output formats for scripting: `--format json` prints structured objects, `--format csv` writes comma-separated rows, and `--quiet` prints only IDs. Combine these with `--cache-dir` to target alternative cache locations. See the [Manage your cache](./manage-cache) guide for advanced workflows.
+The command supports several output formats for scripting: `--format json` prints structured objects, `--format csv` writes comma-separated rows, and `--quiet` prints only IDs. Use `--sort` to order entries by `accessed`, `modified`, `name`, or `size` (append `:asc` or `:desc` to control order), and `--limit` to restrict results to the top N entries. Combine these with `--cache-dir` to target alternative cache locations. See the [Manage your cache](./manage-cache) guide for advanced workflows.
 
 Delete cache entries selected with `hf cache ls --q` by piping the IDs into `hf cache rm`:
 

docs/source/en/guides/manage-cache.md

@@ -401,7 +401,7 @@ Found 2 repo(s) for a total of 2 revision(s) and 3.0G on disk.
 
 Need machine-friendly output? Use `--format json` to get structured objects or
 `--format csv` for spreadsheets. Alternatively `--quiet` prints only identifiers (one
-per line) so you can pipe them into other tooling. Combine these options with
+per line) so you can pipe them into other tooling. Use `--sort` to order entries by `accessed`, `modified`, `name`, or `size` (append `:asc` or `:desc` to control order), and `--limit` to restrict results to the top N entries. Combine these options with
 `--cache-dir` when you need to inspect a cache stored outside of `HF_HOME`.
 
 **Filter with common shell tools**

docs/source/en/package_reference/cli.md

@@ -171,6 +171,8 @@ $ hf cache ls [OPTIONS]
 * `-f, --filter TEXT`: Filter entries (e.g. 'size>1GB', 'type=model', 'accessed>7d'). Can be used multiple times.
 * `--format [table|json|csv]`: Output format.  [default: table]
 * `-q, --quiet`: Print only IDs (repo IDs or revision hashes).
+* `--sort [accessed|accessed:asc|accessed:desc|modified|modified:asc|modified:desc|name|name:asc|name:desc|size|size:asc|size:desc]`: Sort entries by key. Supported keys: 'accessed', 'modified', 'name', 'size'. Append ':asc' or ':desc' to explicitly set the order (e.g., 'modified:asc'). Defaults: 'accessed', 'modified', 'size' default to 'desc' (newest/biggest first); 'name' defaults to 'asc' (alphabetical).
+* `--limit INTEGER`: Limit the number of results returned. Returns only the top N entries after sorting.
 * `--help`: Show this message and exit.
 
 ### `hf cache prune`

src/huggingface_hub/cli/cache.py

@@ -62,6 +62,25 @@ class _DeletionResolution:
 _FILTER_PATTERN = re.compile(r"^(?P<key>[a-zA-Z_]+)\s*(?P<op>==|!=|>=|<=|>|<|=)\s*(?P<value>.+)$")
 _ALLOWED_OPERATORS = {"=", "!=", ">", "<", ">=", "<="}
 _FILTER_KEYS = {"accessed", "modified", "refs", "size", "type"}
+_SORT_KEYS = {"accessed", "modified", "name", "size"}
+_SORT_PATTERN = re.compile(r"^(?P<key>[a-zA-Z_]+)(?::(?P<order>asc|desc))?$")
+_SORT_DEFAULT_ORDER = {
+    # Default ordering: accessed/modified/size are descending (newest/biggest first), name is ascending
+    "accessed": "desc",
+    "modified": "desc",
+    "size": "desc",
+    "name": "asc",
+}
+
+
+# Dynamically generate SortOptions enum from _SORT_KEYS
+_sort_options_dict = {}
+for key in sorted(_SORT_KEYS):
+    _sort_options_dict[key] = key
+    _sort_options_dict[f"{key}_asc"] = f"{key}:asc"
+    _sort_options_dict[f"{key}_desc"] = f"{key}:desc"
+
+SortOptions = Enum("SortOptions", _sort_options_dict, type=str, module=__name__)  # type: ignore
 
 
 @dataclass(frozen=True)
@@ -378,6 +397,60 @@ def _compare_numeric(left: Optional[float], op: str, right: float) -> bool:
     return comparisons[op]
 
 
+def compile_cache_sort(sort_expr: str) -> tuple[Callable[[CacheEntry], tuple[Any, ...]], bool]:
+    """Convert a `hf cache ls` sort expression into a key function for sorting entries.
+
+    Returns:
+        A tuple of (key_function, reverse_flag) where reverse_flag indicates whether
+        to sort in descending order (True) or ascending order (False).
+    """
+    match = _SORT_PATTERN.match(sort_expr.strip().lower())
+    if not match:
+        raise ValueError(f"Invalid sort expression: '{sort_expr}'. Expected format: 'key' or 'key:asc' or 'key:desc'.")
+
+    key = match.group("key").lower()
+    explicit_order = match.group("order")
+
+    if key not in _SORT_KEYS:
+        raise ValueError(f"Unsupported sort key '{key}' in '{sort_expr}'. Must be one of {list(_SORT_KEYS)}.")
+
+    # Use explicit order if provided, otherwise use default for the key
+    order = explicit_order if explicit_order else _SORT_DEFAULT_ORDER[key]
+    reverse = order == "desc"
+
+    def _sort_key(entry: CacheEntry) -> tuple[Any, ...]:
+        repo, revision = entry
+
+        if key == "name":
+            # Sort by cache_id (repo type/id)
+            value: Any = repo.cache_id.lower()
+            return (value,)
+
+        if key == "size":
+            # Use revision size if available, otherwise repo size
+            value = revision.size_on_disk if revision is not None else repo.size_on_disk
+            return (value,)
+
+        if key == "accessed":
+            # For revisions, accessed is not available per-revision, use repo's last_accessed
+            # For repos, use repo's last_accessed
+            value = repo.last_accessed if repo.last_accessed is not None else 0.0
+            return (value,)
+
+        if key == "modified":
+            # Use revision's last_modified if available, otherwise repo's last_modified
+            if revision is not None:
+                value = revision.last_modified if revision.last_modified is not None else 0.0
+            else:
+                value = repo.last_modified if repo.last_modified is not None else 0.0
+            return (value,)
+
+        # Should never reach here due to validation above
+        raise ValueError(f"Unsupported sort key: {key}")
+
+    return _sort_key, reverse
+
+
 def _resolve_deletion_targets(hf_cache_info: HFCacheInfo, targets: list[str]) -> _DeletionResolution:
     """Resolve the deletion targets into a deletion resolution."""
     repo_lookup, revision_lookup = build_cache_index(hf_cache_info)
@@ -458,13 +531,28 @@ def ls(
             help="Print only IDs (repo IDs or revision hashes).",
         ),
     ] = False,
+    sort: Annotated[
+        Optional[SortOptions],
+        typer.Option(
+            help="Sort entries by key. Supported keys: 'accessed', 'modified', 'name', 'size'. "
+            "Append ':asc' or ':desc' to explicitly set the order (e.g., 'modified:asc'). "
+            "Defaults: 'accessed', 'modified', 'size' default to 'desc' (newest/biggest first); "
+            "'name' defaults to 'asc' (alphabetical).",
+        ),
+    ] = None,
+    limit: Annotated[
+        Optional[int],
+        typer.Option(
+            help="Limit the number of results returned. Returns only the top N entries after sorting.",
+        ),
+    ] = None,
 ) -> None:
     """List cached repositories or revisions."""
     try:
         hf_cache_info = scan_cache_dir(cache_dir)
     except CacheNotFound as exc:
         print(f"Cache directory not found: {str(exc.cache_dir)}")
-        raise typer.Exit(code=1)
+        raise typer.Exit(code=1) from exc
 
     filters = filter or []
 
@@ -478,6 +566,20 @@ def ls(
     for fn in filter_fns:
         entries = [entry for entry in entries if fn(entry[0], entry[1], now)]
 
+    # Apply sorting if requested
+    if sort:
+        try:
+            sort_key_fn, reverse = compile_cache_sort(sort.value)
+            entries.sort(key=sort_key_fn, reverse=reverse)
+        except ValueError as exc:
+            raise typer.BadParameter(str(exc)) from exc
+
+    # Apply limit if requested
+    if limit is not None:
+        if limit < 0:
+            raise typer.BadParameter(f"Limit must be a positive integer, got {limit}.")
+        entries = entries[:limit]
+
     if quiet:
         for repo, revision in entries:
             print(revision.commit_hash if revision is not None else repo.cache_id)

tests/test_cli.py

@@ -126,6 +126,31 @@ def test_ls_quiet_revisions(self, runner: CliRunner) -> None:
         assert result.exit_code == 0
         assert result.stdout.strip() == revision.commit_hash
 
+    def test_ls_with_sort(self, runner: CliRunner) -> None:
+        repo1 = _make_repo("user/model1", revisions=[_make_revision("d" * 40)])
+        repo2 = _make_repo("user/model2", revisions=[_make_revision("e" * 40)])
+        repo3 = _make_repo("user/model3", revisions=[_make_revision("f" * 40)])
+        entries = [(repo1, None), (repo2, None), (repo3, None)]
+        repo_refs_map = {repo1: frozenset(), repo2: frozenset(), repo3: frozenset()}
+
+        with (
+            patch("huggingface_hub.cli.cache.scan_cache_dir"),
+            patch(
+                "huggingface_hub.cli.cache.collect_cache_entries",
+                return_value=(entries, repo_refs_map),
+            ),
+        ):
+            result = runner.invoke(app, ["cache", "ls", "--sort", "name:desc", "--limit", "2"])
+
+        assert result.exit_code == 0
+        stdout = result.stdout
+
+        # Check alphabetical order
+        assert stdout.index("model3") < stdout.index("model2")  # descending order
+
+        # Check limit of 2 entries
+        assert "model1" not in stdout
+
     def test_rm_revision_executes_strategy(self, runner: CliRunner) -> None:
         revision = _make_revision("c" * 40)
         repo = _make_repo("user/model", revisions=[revision])