add cookbook for Deepseek v4 pro PD disagg

[yan-lgtm]

Motivation

Modifications

Accuracy Tests

Speed Tests and Profiling

Checklist

• Format your code according to the Format code with pre-commit.
• Add unit tests according to the Run and add unit tests.
• Update documentation according to Write documentations.
• Provide accuracy and speed benchmark results according to Test the accuracy and Benchmark the speed.
• Follow the SGLang code style guidance.

Review and Merge Process

1. Ping Merge Oncalls to start the process. See the PR Merge Process.
2. Get approvals from CODEOWNERS and other reviewers.
3. Trigger CI tests with comments or contact authorized users to do so.
   • Common commands include /tag-and-rerun-ci, /tag-run-ci-label, /rerun-failed-ci
4. After green CI and required approvals, ask Merge Oncalls or people with Write permission to merge the PR.

[gemini-code-assist]

Code Review

This pull request adds documentation and configuration for DeepSeek-V4 PD-Disagg on B200 clusters, including RDMA setup guides and memory optimizations to prevent OOM on 'big' models. Feedback suggests improving the RDMA troubleshooting script by iterating over all InfiniBand devices instead of hardcoding a single device name.

[gemini-code-assist]

The troubleshooting script hardcodes mlx5_0 when checking the pkey table. However, the preceding text mentions that mlx5_7 is the typical default for B200, and the previous script iterates over all mlx5_* devices. Hardcoding mlx5_0 here might lead users to inspect the wrong device. It is better to iterate over all devices to ensure the active one is covered.

for d in /sys/class/infiniband/mlx5_*; do
  echo "--- $(basename $d) ---"
  for i in $(seq 0 7); do
    v=$(cat "$d/ports/1/pkeys/$i" 2>/dev/null)
    [ -n "$v" ] && [ "$v" != "0x0000" ] && echo "pkey[$i]=$v"
  done
done