[Doc] Sync four function docstrings with their signatures (#45437)

Signed-off-by: Daoyuan Li <94409450+DaoyuanLi2816@users.noreply.github.com>
Co-authored-by: Wentao Ye <44945378+yewentao256@users.noreply.github.com>
Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>

vllm/_aiter_ops.py

@@ -2637,7 +2637,8 @@ class rocm_aiter_ops:
 
         Args:
             tensor: The input weight tensor to be shuffled.
-            layout: The block layout to use, defaults to (16, 4).
+            nLane: Number of lanes in the shuffle layout.
+            gate_up: Whether the weight is for w13 (True) or w2 (False).
 
         Returns:
             torch.Tensor: The shuffled tensor.
@@ -2875,7 +2876,6 @@ class rocm_aiter_ops:
             hc_sinkhorn_eps: sinkhorn epsilon
             hc_post_mult_value: post-mix multiplier value
             sinkhorn_repeat: number of sinkhorn iterations
-            n_splits: split-k factor;
 
         Returns:
             post_mix: shape (..., hc_mult), dtype torch.float32

vllm/_custom_ops.py

@@ -1539,7 +1539,11 @@ def scaled_fp4_quant(
     Args:
         input: The input tensor to be quantized to FP4
         input_global_scale: A scalar scaling factor for the entire tensor.
-        use_8x4_sf_layout: Whether to use the 8x4 or 128x4 layout for the scaling
+        is_sf_swizzled_layout: Whether to store the scaling factors in the
+            swizzled layout (default `True`).
+        backend: Quantization kernel backend to dispatch to. For `"trtllm"`
+            backends the 8x4 scale-factor layout is selected for small
+            batches (m <= 32) instead of the 128x4 layout.
         padded_n: Optional padded K dimension. When provided, the quantized
             output and scale tensors are allocated for ``padded_n``
 

vllm/model_executor/layers/quantization/utils/fp8_utils.py

@@ -909,7 +909,7 @@ def w8a8_triton_block_scaled_mm(
         Bs: The per-block quantization scale for `B`.
         block_size: The block size for per-block quantization. It should
         be 2-dim, e.g., [128, 128].
-        output_dytpe: The dtype of the returned tensor.
+        output_dtype: The dtype of the returned tensor.
     Returns:
         torch.Tensor: The result of matmul.
     """