Reproducing Issue

[Jharry502]

Appreciate your splendid work and unselfish code release!

When I attempt to reproduce your code， I encounter the phenomenon shown below：

"I complete the Setup process and run the code "python generate.py --dataset=HumanEval --task=optimize --epoch=0" on two RTX 4090 GPUs (48 GB vRAM per GPU). It seems that it requires 500 epochs to complete this process, and shows 0% Volatile GPU-Util. Also, it takes too long to complete this process, where at least two days are used to complete only 250 epochs. The only output is the "graph_temporary-epoch_datapoints.py" in the "scoreflow_workspace/temp_gene_workflow_file" shown below.

Could you please tell me if I'm doing something wrong? Looking forward to your reply!

@itsrainingdata @YangLing0818 @yinjjiew

[Thiswycf]

'shows 0% Volatile GPU-Util' because it's waiting for workflow(graph) validation. This may mainly depends on the network(API calling), so I make it local LLM. But I still meet some other questions that fine-tuned generator LLM's graph generation validation pass ratio is pretty low.