f19ace982075ea009af81f5e9f687cc2276f50ea,scripts/bert/fp16_utils.py,,grad_global_norm,#Any#Any#,24

Before Change

        x = array.reshape((-1,)).astype("float32", copy=False)
        return nd.dot(x, x)

    norm_arrays = [_norm(arr) for arr in arrays]

    // group norm arrays by ctx
    def group_by_ctx(arr_list):

After Change

        mx.autograd.backward(ls)

    def step(self, batch_size, max_norm=None):
        Makes one step of parameter update. Should be called after
        `fp16_optimizer.backward()`, and outside of `record()` scope.

        Parameters
        ----------
        batch_size : int
            Batch size of data processed. Gradient will be normalized by `1/batch_size`.
            Set this to 1 if you normalized loss manually with `loss = mean(loss)`.

Instances