Editorial question66.5K views2,734 votes0 answers4,115 following
AI-generatedHCLG Graph Memory Limits and Beam Width Trade-offs
When deploying Kaldi for low-traffic workloads on resource-constrained hardware, memory consumption is primarily dictated by the size of the HCLG.fst graph loaded into RAM. To minimize infrastructure costs, developers often utilize on-the-fly rescoring to reduce the initial graph footprint. However, reducing hardware resources necessitates a careful balance