GLM-5.3-Flash support (#36507)

Co-authored-by: zRzRzRzRzRzRzR <Yuxuan.Zhang2@liverpool.ac.uk>
Co-authored-by: Shijin Zhang <75300765+Dovis01@users.noreply.github.com>
Co-authored-by: zanes-ops <zanes@nvidia.com>
Co-authored-by: Baizhou Zhang <sobereddiezhang@gmail.com>
Co-authored-by: Jian Chen <jianchen0311@gmail.com>
Co-authored-by: zijiexia <37504505+zijiexia@users.noreply.github.com>
Co-authored-by: andyluo7 <43718156+andyluo7@users.noreply.github.com>
Co-authored-by: Ehsan Akhgari <ehsan.akhgari@gmail.com>
Co-authored-by: kpham-sgl <khoa.pham@radixark.ai>
Co-authored-by: BBuf <1182563586@qq.com>
Co-authored-by: Raiden Makoto <81530826+Raiden-Makoto@users.noreply.github.com>

test/registered/kernels/ops/attention/test_dsa_indexer.py

@@ -219,6 +219,7 @@ class MockModelRunner:
 
         self.sliding_window_size = None
         self.page_size = self.config["page_size"]
+        self.max_running_requests = max_batch_size
 
         # Create req_to_token_pool
         self.req_to_token_pool = type(
@@ -1240,6 +1241,7 @@ class TestDSAIndexer(CustomTestCase):
                 backend.use_fused_topk = True
                 backend.dsa_topk_backend = topk_backend
                 backend.dsa_index_topk = 2048
+                backend.dsa_index_kpool = 1
                 backend.dsa_decode_impl = "fa3"
                 backend.req_to_token = torch.empty(
                     2, 4096, dtype=torch.int32, device=self.device

test/registered/kernels/ops/attention/test_fused_verify_triton_gdn.py

@@ -26,7 +26,7 @@ try:
 except ImportError:
     KERNELS_AVAILABLE = False
 
-register_cuda_ci(est_time=6, stage="base-b-kernel-unit", runner_config="1-gpu-large")
+register_cuda_ci(est_time=20, stage="base-b-kernel-unit", runner_config="1-gpu-large")
 register_amd_ci(est_time=10, suite="nightly-amd-kernel-1-gpu", nightly=True)
 
 
@@ -234,5 +234,40 @@ def test_mtp_single_step_decode(N: int):
     assert state_fail_rate < 0.01, f"State mismatch: fail_rate={state_fail_rate:.2f}%"
 
 
+@pytest.mark.skipif(not KERNELS_AVAILABLE, reason="Kernels not available")
+def test_verify_scratch_pitch_uses_allocated_steps():
+    # Gear below the allocated step dim must not spill into the neighbor block.
+    N, T, ALLOCATED = 2, 4, 8
+    H, HV, K, V = 16, 32, 128, 128
+
+    A_log, dt_bias, a, b, q, k, v, state, indices, cu_seqlens = _make_tensors(
+        N, T, H, HV, K, V
+    )
+    buffer = torch.full(
+        (N + 1, ALLOCATED, HV, V, K), float("nan"), dtype=torch.float32, device="cuda"
+    )
+
+    run_fused_mtp(
+        A_log,
+        dt_bias,
+        q,
+        k,
+        v,
+        a,
+        b,
+        state,
+        indices,
+        cu_seqlens,
+        disable_state_update=True,
+        intermediate_states_buffer=buffer,
+        intermediate_state_indices=indices,
+        cache_steps=T,
+    )
+
+    assert not torch.isnan(buffer[:N, :T]).any()
+    assert torch.isnan(buffer[N:]).all()
+    assert torch.isnan(buffer[:N, T:]).all()
+
+
 if __name__ == "__main__":
     sys.exit(pytest.main([__file__, "-v", "-s"]))