Add Cohere Command R7B and Command A+ to vLLM extended suite (#243)

[i-kosarev]

• Add Cohere Command R7B and Command A+ to vLLM extended suite

Also bump the pyt_vllm base image to vllm/vllm-openai-rocm:v0.28.0.

Claude-Session: https://claude.ai/code/session_015S3frHBxF3j5AsWL6gt3dk

• Drop comments and keep vision tower for Cohere extended configs

Claude-Session: https://claude.ai/code/session_015S3frHBxF3j5AsWL6gt3dk

---

Motivation

Technical Details

Test Plan

Test Result

Submission Checklist

• Look over the contributing guidelines at https://github.com/ROCm/TheRock/blob/main/GOVERNANCE.md#pull-requests.

[Copilot]

🟡 Changes recommended

The extended configurations use environment mappings that prevent the vLLM server from starting.

Get a fresh assessment by requesting another Copilot review.

Pull request overview

Adds Cohere Command R7B and Command A+ to the vLLM extended suite and updates the ROCm image to v0.28.0.

Changes:

• Registers both Cohere models and benchmark configurations.
• Updates the vLLM Docker image and documentation.
• Adds release-note coverage.

File summaries

File	Summary
scripts/vllm/models.json	Registers the Cohere workloads.
scripts/vllm/configs/extended.yaml	Adds serving configurations; environment mappings at lines 92 and 103 produce invalid shell syntax.
docker/pyt_vllm.ubuntu.amd.Dockerfile	Updates the vLLM base image to v0.28.0.
benchmark/vllm/README.md	Updates benchmark documentation and model listings.

Review details

Suppressed comments (1)

scripts/vllm/configs/extended.yaml:104

• env is parsed as a mapping here, but run_vllm.py interpolates this value directly into the shell command (f"{env} ..."), producing {'VLLM_ROCM_USE_AITER': 1} vllm serve ..., which is not a valid environment assignment and prevents the server from starting. Render the mapping as KEY=VALUE assignments in the runner, or use the string form accepted by the current runner for this entry.

  env:
    VLLM_ROCM_USE_AITER: 1

• Files reviewed: 4/4 changed files
• Comments generated: 1
• Review effort level: Lite

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.