Simplify file_cache logic and re-enable for sequence salience demo.

PiperOrigin-RevId: 638684231

lit_nlp/examples/lm_salience_demo.py

@@ -54,10 +54,7 @@
 from lit_nlp import server_flags
 from lit_nlp.api import layout
 from lit_nlp.examples.datasets import lm as lm_data
-
-# TODO(b/333698148): file_cache doesn't work well with certain HF and KerasNLP
-# preset names. Disabling until resolved.
-# from lit_nlp.lib import file_cache
+from lit_nlp.lib import file_cache
 
 # NOTE: additional flags defined in server_flags.py
 
@@ -282,12 +279,14 @@ def main(argv: Sequence[str]) -> Optional[dev_server.LitServerType]:
     model_name, path = model_string.split(":", 1)
     logging.info("Loading model '%s' from '%s'", model_name, path)
 
-    # TODO(b/333698148): file_cache doesn't work well with certain HF and
-    # KerasNLP preset names. Disabling until resolved.
-    # path = file_cache.cached_path(
-    #     path,
-    #     extract_compressed_file=path.endswith(".tar.gz"),
-    # )
+    # Limit scope of caching to archive files and remote paths, as some preset
+    # names like "google/gemma-1.1-7b-it" look like file paths but should not
+    # be handled as such.
+    if path.endswith(".tar.gz") or file_cache.is_remote(path):
+      path = file_cache.cached_path(
+          path,
+          extract_compressed_file=path.endswith(".tar.gz"),
+      )
 
     if _DL_FRAMEWORK.value == "kerasnlp":
       # pylint: disable=g-import-not-at-top

lit_nlp/lib/file_cache.py

@@ -217,6 +217,7 @@ def filename_fom_url(url: str, etag: Optional[str] = None) -> str:
 
 
 def is_remote(url_of_filepath: str) -> bool:
+  """Check if a path represents a remote URL or non-local file."""
   parsed = urllib_parse.urlparse(url_of_filepath)
   return parsed.scheme in ('http', 'https')