ref(upstream): FULL TREE — Deep-Spark xllm (1470) + ds_vllm csrc/models (703)

Replaces cherry-picked upstream_ref with complete source trees.

xllm/ — Iluvatar official C++ inference engine (15MB, 1470 files)
  Complete: kernels → layers → models → runtime → scheduler → api
  Excluded: .git, binary images, third_party submodule checkouts

ds_vllm/ — Iluvatar official vllm fork (8MB, 703 files)
  Included: csrc/ (ALL CUDA kernels), fused_moe/, qwen3_5 model, _custom_ops
  Excluded: tests, benchmarks, docs, examples (not needed for reference)

Critical call chains now fully traceable:
  MoE: moe_topk_softmax_kernels.cuh → ixformer.h → fused_moe.cpp → layer
  GDN: qwen3_gated_delta_net_base.cpp → qwen3_5_gated_delta_net.cpp
  Attention: ixformer.h → xllm_paged_attention → attention.cpp

upstream_ref/xllm/tools/compare_tensor.py

@@ -0,0 +1,57 @@
+import torch
+import numpy as np
+
+
+def compare_tensors(
+    a: torch.Tensor,
+    b: torch.Tensor,
+    tol: float = 1e-6,
+    verbose: bool = False
+) -> int:
+    """
+    Compare two PyTorch tensors and count the number of elements whose absolute difference
+    exceeds the given tolerance.
+
+    Args:
+        a (torch.Tensor): The first tensor to compare.
+        b (torch.Tensor): The second tensor to compare.
+        tol (float, optional): The absolute tolerance threshold. Defaults to 1e-6.
+        verbose (bool, optional): If True, print the indices and values of differing elements. Defaults to False.
+
+    Returns:
+        int: The number of elements where abs(a - b) > tol.
+
+    Raises:
+        ValueError: If the shapes of the input tensors do not match.
+    """
+    # Check if tensor shapes are the same
+    if a.shape != b.shape:
+        raise ValueError(f"Shape mismatch: {a.shape} vs {b.shape}")
+
+    # Create a boolean mask where differences exceed the tolerance
+    diff_mask = (a - b).abs() > tol
+
+    # Count the number of differing elements
+    diff_count = int(diff_mask.sum().item())
+
+    # If verbose, print details of differing elements
+    if verbose and diff_count > 0:
+        indices = torch.nonzero(diff_mask, as_tuple=False)
+        for idx in indices:
+            i, j = idx[0].item(), idx[1].item()
+            print(
+                f"diff at {i},{j}: "
+                f"{a[i, j].item():.6f} - {b[i, j].item():.6f} = "
+                f"{(a[i, j] - b[i, j]).item():.6f}"
+            )
+
+    return diff_count
+
+
+if __name__ == "__main__":
+    # example:
+    # a = torch.load("/path/to/a.pt")
+    # b = torch.load("/path/to/b.pt")
+    # diff_count = compare_tensors(a, b)
+    # print(f"diff count: {diff_count}")
+    pass