if config.MaxPerCore != nil && *config.MaxPerCore > 0 {
		floor := 0
		if config.Min != nil {
			floor = int(*config.Min)
		}
		scaled := int(*config.MaxPerCore) * detectNumCPU()
		if scaled > floor {
			logger.V(3).Info("GetConntrackMax: using scaled conntrack-max-per-core")
			return scaled, nil
		}
		logger.V(3).Info("GetConntrackMax: using conntrack-min")
		return floor, nil
	}
	return 0, nil
}

MIN_COMBINED, a small bias is introduced where repeated iterations of quantizing
and dequantizing will introduce a larger and larger error.

*SCALED mode Example*

`SCALED` mode matches the quantization approach used in
`QuantizeAndDequantize{V2|V3}`.

If the mode is `SCALED`, the quantization is performed by multiplying each
input value by a scaling_factor.

        auto shape = GetAsVector(quant_tensor->shape());
        if (kUseUpdatedHybridSchemeDefault) {
          EXPECT_EQ(quant_tensor->quantization()->scale()->size(), shape[0]);
        } else {
          EXPECT_EQ(quant_tensor->quantization()->scale()->size(), 1);
        }
      } else {
        EXPECT_EQ(quant_tensor->type(), TensorType_FLOAT32);
      }
    }

    auto scaled = rewriter.create<chlo::BroadcastMulOp>(
        op.getLoc(), result_type, iota, op.getDelta(),
        hlo::getBroadcastDimensionsAttr(&rewriter, iota, op.getDelta()));
    rewriter.replaceOpWithNewOp<chlo::BroadcastAddOp>(
        op, result_type, scaled, op.getStart(),
        hlo::getBroadcastDimensionsAttr(&rewriter, scaled, op.getStart()));
    return success();
  }
};

    float_min = -float_max;
  }

  ASSERT_THAT(quantized_quant_params.scale, SizeIs(1));
  ASSERT_THAT(quantized_quant_params.zero_point, SizeIs(1));
  float scale = (float_max - float_min) / ((1 << bit_num) - 1);
  EXPECT_THAT(scale, FloatNear(quantized_quant_params.scale[0], eps));
}

class QuantizeModelTest : public testing::Test {
 protected:
  QuantizeModelTest() {

	completionMode := completionModeStr(job)
	var diff int

	// Updating succeeded metric must be handled differently
	// for Indexed Jobs to handle the case where the job has
	// been scaled down by reducing completions & parallelism
	// in tandem, and now a previously completed index is
	// now out of range (i.e. index >= spec.Completions).
	if isIndexedJob(job) {
		completions := int(*job.Spec.Completions)

function setup-fluentd {
  local -r dst_dir="$1"
  local -r fluentd_gcp_yaml="${dst_dir}/fluentd-gcp/fluentd-gcp-ds.yaml"
  local -r fluentd_gcp_scaler_yaml="${dst_dir}/fluentd-gcp/scaler-deployment.yaml"
  # Ingest logs against new resources like "k8s_container" and "k8s_node" if
  # LOGGING_STACKDRIVER_RESOURCE_TYPES is "new".
  # Ingest logs against old resources like "gke_container" and "gce_instance" if

- From now on, the HPA controller will return an error for the container resource metrics when the feature gate "HPAContainerMetrics" is disabled. As a result, HPA with a container resource metric performs no scale-down and performs only scale-up based on other metrics. ([#116043](https://github.com/kubernetes/kubernetes/pull/116043), [@sanposhiho](https://github.com/sanposhiho)) [SIG API Machinery, Apps and Autoscaling]

- Fixed a `v1.28.0` regression where `kube-controller-manager` can crash when `StatefulSet` with `Parallel` policy and PVC labels are scaled up. ([#121142](https://github.com/kubernetes/kubernetes/pull/121142), [@aleksandra-malinowska](https://github.com/aleksandra-malinowska))

`Policy` optionally took in a `tf.compat.v1.mixed_precision.LossScale` in the constructor, which defaulted to a dynamic loss scale for the `"mixed_float16"` policy and no loss scale for other policies. In `Model.compile`, if the model's policy had a loss scale, the optimizer would be wrapped with a `LossScaleOptimizer`. With the non-experimental `Policy`, there is no loss scale associated with the `Policy`, and `Model.compile` wraps the optimizer with a `LossScaleOptimizer` if and only if the policy...