← Workflow
Optional · Step 0 · Quick check
Parallelism Advisor
Recommend TP, PP, EP, and DP strategies.
Results
- Tensor parallel (TP)
- 1
- Pipeline parallel (PP)
- 32
- Expert parallel (EP)
- 1
- Data parallel (DP)
- 1
- Note
- TP=1 for weight sharding; PP if layers exceed single-node memory.
Optional step 0 — continue to Define workload (Lens) for the full platform workflow.