Install any skill in seconds. Free to start, no credit card required.
Get Started Free →yangge521/gpu-concurrent
多 GPU 并发推理管理,支持 CUDA_VISIBLE_DEVICES/MUSA_VISIBLE_DEVICES 环境变量隔离与进程管理
Passed 19 of 22 test cases with the skill, and 15 of 22 without it.
and 1 more in the full report
4 fixed15 passed either way3 still failing0 broke
Tokens, turns and time describe resource use, not a dollar price.
One of 4 test cases the skill fixed. Same model, same prompt, with and without the skill.
To query GPU utilization and memory consumption on NVIDIA we run nvidia-smi. What is the equivalent command-line tool used for querying metrics on Moore Threads GPUs?
| Checked | Without skill | With skill |
|---|---|---|
| Names mthreads-gmi as the query tool for Moore Threads GPUs | Missed | Passed |
The instructions your model reads when the skill loads.
管理多 GPU 环境下的并发推理任务,通过环境变量隔离 GPU 资源,实现每张 GPU 独立运行推理进程。已在 8x MTT S4000 集群上验证。
| 测试模式 | 总 FPS | 达标路数 | 状态 |
|---|---|---|---|
| 8卡 x 1路 | 374.84 | 16/16 | 全部达标 |
| 8卡 x 2路 | 493.15 | 12/16 | 4路未达标 |
NVIDIA 使用 CUDA_VISIBLE_DEVICES=0,1 指定 GPU。摩尔线程使用 MUSA_VISIBLE_DEVICES=0,1。在启动推理进程前设置环境变量。
使用 subprocess.Popen 启动独立进程,每个进程通过 env 参数绑定特定 GPU。按 NUM_GPUS x ROUTES_PER_GPU 计算总路数。
test cases passed, vs 15 / 22 without
More skills from the same source and nearby in the registry, with their headline benchmark lift.