Troubleshooting

◈CUDA Out of MemoryTraining crashes with memory errors ⏱Training Too SlowLow throughput, GPU underutilized ▲Loss is NaNLoss becomes NaN and training breaks ▽Loss Not DecreasingTraining loss stays flat or increases ▣️GPU Not Detectedtorch.cuda.is_available() returns False ◇Dataset ErrorData loading fails or produces bad batches ▦Checkpoint ErrorCannot save or load checkpoints ⬡Distributed Training FailureNCCL errors, scaling issues ◐Poor EvaluationModel scores lower than expected ○Model Outputs Are BadGenerated text is incoherent or off-target