chore: fix some typos (#18577)

Co-authored-by: Liangsheng Yin <lsyincs@gmail.com>

python/sglang/srt/layers/quantization/fp8_kernel.py

@@ -219,7 +219,7 @@ def _per_token_group_quant_8bit_raw(
     quantized tensor along with the scaling factor used for quantization.
 
     Args:
-        x: The input tenosr with ndim >= 2.
+        x: The input tensor with ndim >= 2.
         group_size: The group size used for quantization.
         eps: The minimum to avoid dividing zero.
         dtype: The dype of output tensor.
@@ -635,7 +635,7 @@ def static_quant_fp8(
     quantized tensor along with the scaling factor used for quantization.
 
     Args:
-        x: The input tenosr with ndim >= 2.
+        x: The input tensor with ndim >= 2.
         x_s: The quantization scale.
         repeat_scale: Whether to broadcast per-tensor scale to per-channel scale.
         dtype: The dype of output tensor.

python/sglang/srt/layers/quantization/int8_kernel.py

@@ -143,7 +143,7 @@ def per_token_group_quant_int8(
     quantized tensor along with the scaling factor used for quantization.
 
     Args:
-        x: The input tenosr with ndim >= 2.
+        x: The input tensor with ndim >= 2.
         group_size: The group size used for quantization.
         eps: The minimum to avoid dividing zero.
         dtype: The dype of output tensor. Note that only `torch.int8` is supported for now.

python/sglang/srt/layers/quantization/quark/utils.py

@@ -128,10 +128,10 @@ def b_dynamic_mxfp4_quant(x):
     return x.view(h, b, d // 2), x_scales.view(h, b, d // 32)
 
 
-def mxfp4_to_f32(x, is_threed):
+def mxfp4_to_f32(x, is_3d):
     # 2 because we pack fp4 in uint8.
     x = x.repeat_interleave(2, dim=-1)
-    if is_threed:
+    if is_3d:
         x[..., ::2] = x[..., ::2] & 0xF
         x[..., 1::2] = x[..., 1::2] >> 4
     else: