Bad Request, code: 400

[lihan97]

When executing example.sh in the ./apiexample directory, the following error consistently occurs: begin request Namespace(data_path='./data/baron_human_samp_19264_fromsaver_demo.csv', input_type='singlecell', output_type='cell', pool_type='all', pre_normalized='F', save_path='./demo/Baron_demo/a5/', tgthighres='a5', token='biomapxxxxxxxxxxxxx', url='https://api.biomap.com/inference/xtrimogene', version='0.2') https://api.biomap.com/inference/xtrimogene {'Accept-Encoding': 'gzip,deflate,br', 'B-Authorization': 'biomapxxxxxxxxxxxx'} code:400 text:{"error":"Failed to process the request(s) for model instance 'xtrimogene_0', message: TypeError: init(): incompatible constructor arguments. The following argument types are supported:\n 1. c_python_backend_utils.InferenceResponse(output_tensors: List[c_python_backend_utils.Tensor], error: c_python_backend_utils.TritonError = None)\n\nInvoked with: kwargs: error=<c_python_backend_utils.TritonError object at 0x7f9049781cf0>\n\nAt:\n /work/xtrimogene/service/xtrimogene/1/model.py(145): execute\n"} reason:Bad Request

[lihan97]

I found this script working properly today. But shortly thereafter, it raised the following error: begin request Namespace(data_path='./data/gene_batch.npy', input_type='singlecell', output_type='gene', pool_type='all', pre_normalized='A', save_path='./demo/GEARS_demo_batch/f1/', tgthighres='f1', token='biomapxxxxxxxxxx', url='https://api.biomap.com/inference/xtrimogene', version='0.1') https://api.biomap.com/inference/xtrimogene {'Accept-Encoding': 'gzip,deflate,br', 'B-Authorization': 'biomapxxxxxxxxxxxxxxx'} code:400 text:{"error":"Failed to process the request(s) for model instance 'xtrimogene_0', message: RuntimeError: CUDA error: device-side assert triggered\nCUDA kernel errors might be asynchronously reported at some other API call,so the stacktrace below might be incorrect.\nFor debugging consider passing CUDA_LAUNCH_BLOCKING=1.\n\nAt:\n /opt/conda/lib/python3.7/site-packages/torch/cuda/memory.py(114): empty_cache\n /work/xtrimogene/service/xtrimogene/1/model.py(73): execute\n"} reason:Bad Request

[WhirlFirst]

Hi lihan97, could you please give more information about what kind of data you used? Did you align the gene name and remove all abnormal gene expression data from the input?

[lihan97]

I checked the input data format and ensured its consistency with the provided example data, successfully resolving the issue. Thank you for your suggestion!

I have an additional question: I want to obtain cell/gene embeddings for some downstream tasks. Does the selection of F/T/A for "--pre_normalized" influence the resulting embeddings? If so, which option is recommended?

[WhirlFirst]

Very happy to hear your successful inference!

Yes, this argument is decided by the input data format. For more details, plz refer to the readme in the apiexample folder. https://github.com/biomap-research/scFoundation/tree/main/apiexample#:~:text=is%20singlecell.-,pre_normalized,-%3A%20Controls%20the%20computation

When input_type is singlecell, T or F indicates if the input gene expression data is already normalized+log1p. A means data is normalized+log1p with the total count appended at the end, resulting in a data shape of N*19265. This mode is used for the GEARS task. For bulk input type, F means the T and S token values are log10(sum of gene expression), while T means they are the sum without log transformation. This is useful for bulk data with few sequenced genes.