Everyone understands the pain points of large model training iterations—data often starts at 10TB or more, and each update requires re-uploading everything. This process consumes a lot of time and storage costs.

Walrus has recently optimized this issue. The core improvement is the slice-level incremental update feature—only upload the changed data blocks, while keeping the rest unchanged. It sounds simple, but the results are indeed significant. In a real case, a 10TB training dataset iteration using this solution reduced the time from several hours to just 15 minutes. The cost savings are also substantial, with storage expenses reduced by 70%.

For small and medium-sized AI companies, this solution is particularly practical. It saves time, significantly reduces operational costs, improves data iteration efficiency, and lightens storage burdens. It seems like a great choice.