Update tokenizer path in README for parakeet model (#16495)

Author: mergennachin

backends/apple/metal/metal_backend.py

@@ -32,6 +32,7 @@ def get_device_name(cls) -> str:
     def get_supported_fallback_kernels(cls) -> Dict[str, Any]:
         return {
             "aoti_torch_mps_addmm_out": None,
+            "aoti_torch_mps_bmm_out": None,
             "aoti_torch_mps_convolution": None,
             "aoti_torch_mps_mm_out": None,
             "at::_ops::_scaled_dot_product_attention_math_for_mps::call": None,

backends/apple/metal/runtime/shims/et_metal_ops.h

@@ -27,6 +27,28 @@ AOTITorchError aoti_torch_mps_mm_out(
     AOTITensorHandle self,
     AOTITensorHandle mat2);
 
+/**
+ * ExecutorTorch implementation of aoti_torch_mps_addmm_out.
+ * Performs: out = beta * input + alpha * (mat1 @ mat2)
+ */
+AOTITorchError aoti_torch_mps_addmm_out(
+    AOTITensorHandle out,
+    AOTITensorHandle input,
+    AOTITensorHandle mat1,
+    AOTITensorHandle mat2,
+    int64_t beta,
+    int64_t alpha);
+
+/**
+ * ExecutorTorch implementation of aoti_torch_mps_bmm_out.
+ * Performs batched matrix multiplication: out = self @ mat2
+ * All tensors must be 3-D with matching batch dimensions.
+ */
+AOTITorchError aoti_torch_mps_bmm_out(
+    AOTITensorHandle out,
+    AOTITensorHandle self,
+    AOTITensorHandle mat2);
+
 /**
  * ExecutorTorch implementation of aoti_torch_mps_convolution.
  * Performs 2D convolution operation - matches PyTorch AOTI signature