longcat: a cache hit follows the pipeline's device when none is given

encode_prompt's signature defaults device to None. Upstream's own callers always
pass it, so this never fires today, but a cached tuple returned on the CPU would
meet a DiT on the card — and the fallback is one expression, where discovering it
later is a device-mismatch traceback from inside a generation.

Falls back the way _get_t5_prompt_embeds does.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

tests/test_longcat_prompt_cache.py

@@ -369,6 +369,15 @@ def test_a_lazily_realised_quantised_encoder_is_never_swapped(SVC):
     assert "_encoder_is_quantised(pipe)" in wrap
 
 
+def test_a_hit_without_an_explicit_device_follows_the_pipeline(SVC):
+    """encode_prompt's signature defaults device to None. A cached tuple left on the
+    CPU would then meet a DiT on the card."""
+    wrap = SERVICE.read_text(encoding="utf-8")
+    wrap = wrap[wrap.index("def _wrap_encode_prompt("):]
+    wrap = wrap[:wrap.index("\n@contextlib")]
+    assert 'kwargs.get("device") or getattr(pipe, "device", None)' in wrap
+
+
 def test_the_disk_cache_is_consulted_before_the_encoder_is_touched(SVC):
     """Order matters: in-memory, then disk, then encode. Realising the encoder before
     the disk lookup would spend the 11 GB this feature exists to save."""

tools/longcat_service.py

@@ -683,7 +683,11 @@ def _wrap_encode_prompt(pipe, swap: bool, cache=None, lazy=None):
                 log(f"prompt cache: cannot key this call ({type(exc).__name__}: {exc})")
                 disk_key = None
         if disk_key is not None:
-            hit = cache.get(disk_key, device=kwargs.get("device"),
+            # Upstream's callers always pass device, but its signature defaults it to
+            # None, and a cached tuple left on the CPU would then meet a DiT on the
+            # card. Fall back the way _get_t5_prompt_embeds would.
+            hit = cache.get(disk_key,
+                            device=kwargs.get("device") or getattr(pipe, "device", None),
                             dtype=kwargs.get("dtype"))
             if hit is not None:
                 log("prompt cache hit — the text encoder is not needed for this request")