Qwen3模型PD分离的工作负载仿真结果不正确的问题
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 1.2k
- Forks
- 184
- Avg merge
- 13h 4m
- Merged PRs (30d)
- 1
Description
您好，我在使用新代码做仿真时遇见仿真结果不正确的情况，向您请教。我的使用如下：
首先生成Qwen3-moe模型的prefill工作负载，指令如下：sh ./scripts/inference_workload_with_aiob.sh -m qwen3-235B
-c ./scripts/inference_configs/qwen3_moe_default.json
-s 32768 -b 4 -w 12288 -t 8 -l 16 -e 32 -p prefill
生成的负载文件为：Qwen3-Moe-235B-world_size12288-tp8-pp16-ep32-bs4-seq32768-prefill.txt。
下一步进行analytical仿真，命令如下： ./bin/SimAI_analytical -w aicb/results/workload/Qwen3-Moe-235B-world_size12288-tp8-pp16-ep32-bs4-seq32768-prefill.txt -g 12288 -g_p_s 8 -r prefill- -nv 380 -nic 50
生成的结果中bubble time为100%，其他策略的通信时间占比都为0，请问这是为什么以及如何解决？
感谢感谢。
Contributor guide
No contributing guide indexed for this repository
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Reproduce the case with scripts/inference_workload_with_aiob.sh, scripts/inference_configs/qwen3_moe_default.json, and bin/SimAI_analytical using the commands in the report. Inspect how the generated Qwen3-Moe workload is consumed and how bubble and communication percentages are calculated; done means the cause is identified and the same workload no longer reports 100% bubble time with zero communication time.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python, shell
- Domain
- distributed-systems, tooling
- Issue type
- Bug
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Stale
- Clarity
- Mostly clear
- Newbie friendliness
- 35/100