[verified] feat: complete Ralph task workstreams

tests/test_real_model_backend.py

@@ -24,6 +24,7 @@ from meshnet_node.model_backend import (
     _should_partial_materialize_shard,
     _decoder_attention_mask,
     _int_tensor_header,
+    _tensor_from_bfloat16_bytes,
     _torch_cuda_is_executable,
     build_quantization_config,
     validate_quantization,
@@ -538,6 +539,20 @@ def test_int_tensor_header_serializes_torch_tensors():
     assert header.startswith("1,3:")
 
 
+def test_bfloat16_wire_decode_views_owned_bytes_without_float32_round_trip():
+    """Activation decode stays bf16 and does not clone bytes into bytearray.
+
+    Tags: model, performance, wire
+    """
+    torch = pytest.importorskip("torch")
+    body = torch.tensor([[1, 2]], dtype=torch.bfloat16).view(torch.uint8).numpy().tobytes()
+
+    decoded = _tensor_from_bfloat16_bytes(body, [1, 2], torch)
+
+    assert decoded.dtype == torch.bfloat16
+    assert decoded.tolist() == [[1.0, 2.0]]
+
+
 def test_decoder_attention_mask_is_causal_float_mask():
     "Decoder attention mask is causal float mask\n\nTags: model, node, real-inference"
     torch = pytest.importorskip("torch")