[NVIDIA] Support nvfp4 quantization (#12784)

csrc/torch_bindings.cpp

@@ -423,6 +423,12 @@ TORCH_LIBRARY_EXPAND(TORCH_EXTENSION_NAME, ops) {
   ops.impl("dynamic_per_token_scaled_fp8_quant", torch::kCUDA,
            &dynamic_per_token_scaled_fp8_quant);
 
+  // Compute NVFP4 block quantized tensor.
+  ops.def(
+      "scaled_fp4_quant(Tensor! output, Tensor input,"
+      "                 Tensor! output_scale, Tensor input_scale) -> ()");
+  ops.impl("scaled_fp4_quant", torch::kCUDA, &scaled_fp4_quant);
+
   // Compute int8 quantized tensor for given scaling factor.
   ops.def(
       "static_scaled_int8_quant(Tensor! result, Tensor input, Tensor scale,"