DeepSpeed

6dea5f64 - Fix formatting for benchmarks/opsd benchmark scripts

Fix formatting for benchmarks/opsd benchmark scripts - Add SPDX-License-Identifier and DeepSpeed Team headers to bench_14b_rollout.py, bench_autotp_gc.py, bench_vllm_tp2.py - Remove unused 'sys' import from bench_14b_rollout.py (flake8 F401) - Apply yapf formatting (column_limit=119) - Add #ignore-cuda comments to torch.cuda.* calls in benchmark scripts (these are CUDA-specific benchmarks) Signed-off-by: GitHub <noreply@github.com>