dropp baes64 use binary

tests/test_kv_cache_distributed.py

@@ -278,6 +278,43 @@ def test_session_is_stable_and_decode_payloads_are_single_token():
     assert head_backend.released == [session_id]
 
 
+def test_large_prefill_activation_survives_zstd_compressed_hop():
+    """A prefill body above _COMPRESS_MIN_BYTES travels the hop zstd-compressed.
+
+    The head compresses and sets X-Meshnet-Encoding; the tail's /forward must
+    decompress before shape validation, so a passing generation proves the
+    compressed round trip (a mishandled encoding fails validation with 400).
+    """
+
+    class _BigHeadBackend(_CachedHeadBackend):
+        def encode_prompt(self, prompt, session_id=None):
+            self.prefills.append(session_id)
+            if session_id:
+                self._seq[session_id] = 2048
+            return TensorPayload(
+                body=b"\x00" * (1 * 2048 * 32 * 2),  # 128 KiB, above the zstd threshold
+                shape=[1, 2048, 32],
+                attention_mask_header=None,
+                position_ids_header=None,
+            )
+
+    head_backend = _BigHeadBackend()
+    tail_backend = _CachedTailBackend([(" a", 1), (" b", 2)])
+    head = TorchNodeServer(backend=head_backend, tracker_mode=True)
+    tail = TorchNodeServer(backend=tail_backend)
+    head_port = head.start()
+    tail_port = tail.start()
+    try:
+        content = _chat_once(head_port, tail_port, max_tokens=2)
+    finally:
+        head.stop()
+        tail.stop()
+
+    assert content == " a b"
+    assert tail_backend.calls[0]["mode"] == "prefill"
+    assert tail_backend.calls[0]["shape"] == [1, 2048, 32]
+
+
 def test_eos_token_id_stops_generation():
     head_backend = _CachedHeadBackend()
     tail_backend = _CachedTailBackend([(" a", 1), ("", 99)])