[Intel GPU] DeepSeek V4 5/N: Use sgl-kernel implementation of fused_q_indexer_rope_hadamard_quant to run on XPU (#27873)

Signed-off-by: P V R K Jyothendra Varma <polisettyvarma@gmail.com>
2026-07-14 09:20:50 +08:00
parent ee464fedc6
commit e489685509
1 changed files with 12 additions and 0 deletions
@@ -171,6 +171,18 @@ def fused_q_indexer_rope_hadamard_quant(
            freqs_real,
            positions,
        )
+    elif _is_xpu:
+        from sgl_kernel import fused_q_indexer_rope_hadamard_quant
+
+        fused_q_indexer_rope_hadamard_quant(
+            q_input,
+            q_fp8,
+            weight,
+            weights_out,
+            float(weight_scale),
+            freqs_real,
+            positions,
+        )
    else:
        module = _jit_main_q_indexer_rope_hadamard_quant_module(q_input.dtype)
        module.forward(