accessible from workers. If the worker fails to read from the `work_dir`, it
    falls back to using RPC for dataset graph transfer.
*   Adds support for a new "distributed_epoch" processing mode. This processing
    mode distributes a dataset across all tf.data workers, instead of having
    each worker process the full dataset. See

In some cases, for pure data validation and processing, you can get performance improvements of **20x** or more. This means 2,000% or more. 🤯

// d := bLimbs[i] // T[n+i] = addMulVVW(T[i:n+i], aLimbs, d) // } // // where d is a digit of the multiplier, T[i:n+i] is the shifted // position of the product of that digit, and T[n+i] is the final carry. // Note that T[i] isn't modified after processing the i-th digit. // // Instead of running two loops, one for Step 1 and one for Steps 2–6, // the result of Step 1 is computed during the next loop. This is // possible because each iteration only uses T[i] in Step 2 and then // discards it in Step...

// d := bLimbs[i] // T[n+i] = addMulVVW(T[i:n+i], aLimbs, d) // } // // where d is a digit of the multiplier, T[i:n+i] is the shifted // position of the product of that digit, and T[n+i] is the final carry. // Note that T[i] isn't modified after processing the i-th digit. // // Instead of running two loops, one for Step 1 and one for Steps 2–6, // the result of Step 1 is computed during the next loop. This is // possible because each iteration only uses T[i] in Step 2 and then // discards it in Step...