canergen
commented
1 month ago Hi, it's hard for me to test except our CI tests for Windows in GitHub. Let's try to reduce the reasons for failure. Can you provide whether just the below raises the issue or the train runner:
model_cls = scvi.model.SCVI
model_cls.setup_anndata(adata, categorical_covariate_keys = ['Sample'],
continuous_covariate_keys=['pct_counts_mt'])
tuner = ModelTuner(model_cls)
search_space = {
"n_hidden": tune.choice([92, 128, 192, 256]),
"n_latent": tune.choice([10, 20, 30, 40, 50, 60,70,80]),
"n_layers": tune.choice([1, 2, 3]),
"lr": tune.loguniform(1e-4, 1e-2),
"gene_likelihood": tune.choice(["nb", "zinb"])}I would then disable the GPU. Can you train a model using GPU only in the same environment? Does the same code run in a Colab environment? The error code tells us that there is a wrong parameter somewhere in the code. Can you check: https://github.com/scverse/scvi-tools/issues/2908? We will release a patch fix in the coming days.
GLking123
Pinolinoo
Hi
The running code is as follows：
`model_cls = scvi.model.SCVI model_cls.setup_anndata(adata, categorical_covariate_keys = ['Sample'], continuous_covariate_keys=['pct_counts_mt'])
tuner = ModelTuner(model_cls)
search_space = { "n_hidden": tune.choice([92, 128, 192, 256]), "n_latent": tune.choice([10, 20, 30, 40, 50, 60,70,80]), "n_layers": tune.choice([1, 2, 3]), "lr": tune.loguniform(1e-4, 1e-2), "gene_likelihood": tune.choice(["nb", "zinb"])}
results = tuner.fit( adata, metric="validation_loss", search_space=search_space, num_samples=30， max_epochs=20, resources={"cpu": 10, "gpu": 1}, ) `
The following error message is displayed：
symbolize_win32.inc : 53] raw: syminitialize() failed: 87 fatal python error: abortedFor the above question, could you provide some debugging suggestions? Thank you for your valuable time and assistance. I sincerely look forward to your response！