This commit is contained in:
dean
2026-04-02 21:17:01 +09:00
parent f98f6cb038
commit f00296cf2c
3 changed files with 20 additions and 2 deletions

View File

@@ -56,6 +56,13 @@ public class DockerTrainService {
@Value("${spring.profiles.active}")
private String profile;
@Value("${hyper.parameter.gpus}")
private String hyperGpus;
@Value("${hyper.parameter.gpu-ids}")
private String hyperGpuIds;
private final ModelTrainJobCoreService modelTrainJobCoreService;
/**
@@ -285,11 +292,13 @@ public class DockerTrainService {
// addArg(c, "--gpu-ids", req.getGpuIds()); // null
if ("prod".equals(profile)) {
addArg(c, "--batch-size", 2); // 학습서버 GPU 1개인 곳은 batch-size:2 까지만 가능
addArg(c, "--gpus", "1"); // 학습서버 GPU 1개인 곳은 1이어야 함
addArg(c, "--gpu-ids", "0"); // 학습서버 GPU 1개인 곳은 0이어야 함
} else {
addArg(c, "--batch-size", req.getBatchSize()); // 학습서버 GPU 1개인 곳은 batch-size:2 까지만 가능
}
addArg(c, "--gpus", hyperGpus); // 학습서버 GPU 1개인 곳은 1이어야 함
addArg(c, "--gpu-ids", hyperGpuIds); // 학습서버 GPU 1개인 곳은 0이어야 함
addArg(c, "--lr", req.getLearningRate());
addArg(c, "--backbone", req.getBackbone());
addArg(c, "--epochs", req.getEpochs());

View File

@@ -41,3 +41,7 @@ train:
container_prefix: kamco-cd-train
shm_size: 16g
ipc_host: true
hyper:
parameter:
gpus: 4
gpu-ids: 0,1,2,3

View File

@@ -78,3 +78,8 @@ management:
exposure:
include:
- "health"
hyper:
parameter:
gpus: 1
gpu-ids: 0