build(speechlm2): add cut-cross-entropy to compiled extras

Original-Commit: 2c645f4
Signed-off-by: Piotr Żelasko <pzelasko@nvidia.com>

Author: pzelasko

pyproject.toml

@@ -245,6 +245,7 @@ cu13 = [
 compiled = [
     "onnx-ir==0.2.1",
     "onnxscript==0.7.0",
+    "cut-cross-entropy==25.3.2",
     "deep_ep==1.2.1",
     "nv-grouped-gemm==1.1.4.post8",
     "causal-conv1d==1.6.2.post1",
@@ -258,6 +259,7 @@ compiled = [
 compiled-a100 = [
     "onnx-ir==0.2.1",
     "onnxscript==0.7.0",
+    "cut-cross-entropy==25.3.2",
     "nv-grouped-gemm==1.1.4.post8",
     "causal-conv1d==1.6.2.post1",
     "mamba-ssm==2.3.2.post1",
@@ -475,6 +477,7 @@ deep-ep = ["setuptools"]
 # metadata is mirrored below so this repository controls the CUDA wheel index.
 [tool.uv.sources]
 nemo_automodel = { git = "https://github.com/NVIDIA-NeMo/Automodel.git", rev = "7b15c9bfbc620bec4dd3b819fce2e48726a7ddeb" }
+cut-cross-entropy = { git = "https://github.com/apple/ml-cross-entropy.git", rev = "87a86aba72cfd2f0d8abecaf81c13c4528ea07d8" }
 megatron-fsdp = { git = "https://github.com/yuhezhang-ai/Megatron-LM.git", rev = "455389c480af6b3acdca74c7830c68b3274eb083", subdirectory = "megatron/core/distributed/fsdp/src" }
 deep_ep = { git = "https://github.com/deepseek-ai/DeepEP.git", rev = "7febc6e25660af0f54d95dd781ecdcd62265ecca" }
 torch = [