chore: cleanup garbage code (#29770)

sgl-kernel/benchmark/bench_es_fp8_blockwise_grouped_gemm.py

@@ -79,9 +79,6 @@ def bench_es(
     b_tensors = []
     a_scales_tensors = []
     b_scales_tensors = []
-    if False:
-        print("Token Distributtion: ", group_ms[0:num_groups])
-        print("Token Count: ", sum(group_ms[0:num_groups]))
     for g in range(num_groups):
         m_g = group_ms[g]
         expert_offsets[g + 1] = expert_offsets[g] + m_g

sgl-kernel/python/sgl_kernel/metal.py

@@ -32,19 +32,19 @@ else:
 
 
 def rope_pool_fused(
-    q: "mx.array",
-    k: "mx.array",
-    v: "mx.array",
-    positions: "mx.array",
-    slots: "mx.array",
-    k_pool: "mx.array",
-    v_pool: "mx.array",
+    q: mx.array,
+    k: mx.array,
+    v: mx.array,
+    positions: mx.array,
+    slots: mx.array,
+    k_pool: mx.array,
+    v_pool: mx.array,
     *,
     head_dim: int,
     num_qo_heads: int,
     num_kv_heads: int,
     rope_base: float,
-) -> tuple["mx.array", "mx.array", "mx.array", "mx.array"]:
+) -> tuple[mx.array, mx.array, mx.array, mx.array]:
     """Apply NeoX RoPE to Q/K and scatter K/V into the MLX KV pool.
 
     Args:

sgl-kernel/tests/test_dsv4_norm_rope.py

@@ -1,7 +1,5 @@
 """Tests for DeepSeek-V4 fused norm + RoPE kernels."""
 
-import math
-
 import pytest
 import sgl_kernel
 import torch
@@ -93,7 +91,7 @@ def test_fused_q_norm_rope_preallocated_output():
 
 @pytest.mark.parametrize("batch_size", [1, 8])
 def test_fused_q_indexer_rope_hadamard_quant_runs(batch_size):
-    """Smoke test: kernel runs without errors and produces finite results."""
+    """Basic launch coverage with finite output checks."""
     torch.manual_seed(42)
     num_heads = 4
     head_dim = 128