(xLimbs[i] ^ dLimbs[i])) xLimbs[i], carry = bits.Add(l, l, carry) dLimbs[i], borrow = bits.Sub(xLimbs[i], mLimbs[i], borrow) } // Like in maybeSubtractModulus, we need the subtraction if either it // didn't underflow (meaning 2x + b > m) or if computing 2x + b // overflowed (meaning 2x + b > 2^_W*n > m). needSubtraction = not(choice(borrow)) | choice(carry) } return x.assign(needSubtraction, d) } // Mod calculates out = x mod m. // // This works regardless how large the value of x is. // // The output...

(xLimbs[i] ^ dLimbs[i])) xLimbs[i], carry = bits.Add(l, l, carry) dLimbs[i], borrow = bits.Sub(xLimbs[i], mLimbs[i], borrow) } // Like in maybeSubtractModulus, we need the subtraction if either it // didn't underflow (meaning 2x + b > m) or if computing 2x + b // overflowed (meaning 2x + b > 2^_W*n > m). needSubtraction = not(choice(borrow)) | choice(carry) } return x.assign(needSubtraction, d) } // Mod calculates out = x mod m. // // This works regardless how large the value of x is. // // The output...

*   Added warmup capabilities to `tf.keras.optimizers.schedules.CosineDecay` learning rate scheduler. You can now specify an initial and target learning rate, and our scheduler will perform a linear interpolation between the two after which it will begin a decay phase.

    * If you depended on that previous behavior, you might need to update your code. As always, make sure your tests pass before merging the upgrade.

## 0.90.1

### Upgrades

* ⬆️ Upgrade Starlette range to allow 0.23.1. PR [#5980](https://github.com/tiangolo/fastapi/pull/5980) by [@tiangolo](https://github.com/tiangolo).

### Docs