Closed Yash685 closed 5 months ago
Thank you for your issue. "PIPELINE_NUM" represents the number of batches contained in each super-batch in super-batch pipelining. "CACHE_RATE" represents the proportion of hot vertices computed by the CPU to all vertices in the initial state. "CACHE" indicates whether GPU memory caching of vertex features is used, and "FEATURE_CACHE_RATE" represents the number of vertex features cached by the GPU in the initial state. All other parameters have been deprecated.
Regarding the algorithm differences, "GS_SAMPLE_ALL_GPU" utilizes GPU exclusively for the sampling-gathering-training process, "GS_SAMPLE_PDCACHE" offloads hot vertices computations to the CPU, and "GS_SAMPLE_CACHE" is a backup code.
I apologize for the inconvenience caused, and I will add parameter explanations in future code updates and strive to remove redundant code and deprecated parameters.
UP_DEGREE:1 PIPELINE_NUM:4 CACHE_RATE:0.10 CACHE:1 FEATURE_CACHE_RATE:0.1 PROC_OVERLAP:0 PROC_LOCAL:0 PROC_CUDA:0 PROC_REP:0 PUSHDOWN:0 LOCK_FREE:1
Can you explain to me what these flags mean when specifying in cfg file? Also what are differences between GS_SAMPLE_ALL_GPU, GS_SAMPLE_PDCACHE, GS_SAMPLE_CACHE?