Alexander Zielenski <******@****.***> 1697824555 -0700

)

type JwksFetchMode int

const (
	// Istiod is used to indicate Istiod ALWAYS fetches the JWKs server
	Istiod JwksFetchMode = iota

	// Hybrid is used to indicate Envoy fetches the JWKs server when there is a cluster entry,
	// otherwise fallback to Istiod
	Hybrid

	// Envoy is used to indicate Envoy ALWAYS fetches the JWKs server
	Envoy
)

// String converts JwksFetchMode to readable string.

  QUANTIZED_UINT8 = 3,
  HYBRID = 4
};

inline InferenceType GetInferenceTypeEnum(llvm::StringRef inference_type_str) {
  if (inference_type_str == "FLOAT") {
    return FLOAT;
  } else if (inference_type_str == "QUANTIZED_INT8") {
    return QUANTIZED_INT8;
  } else if (inference_type_str == "QUANTIZED_UINT8") {
    return QUANTIZED_UINT8;
  } else if (inference_type_str == "HYBRID") {
    return HYBRID;
  } else {

area: security
releaseNotes:
  - |
    **Fixed** an issue where gateways cannot fetch JWKS from `jwksUri` in RequestAuthentication 

      GetTransferredElementCount(from_graph, to_graph);

  if (from_inference_type == FLOAT || from_inference_type == HYBRID) {
    // FLOAT <-> HYBRID will have no quant/dequant as well.
    if (to_inference_type == FLOAT || to_inference_type == HYBRID) {
      return 0;
    } else if (to_inference_type == QUANTIZED_INT8 ||
               to_inference_type == QUANTIZED_UINT8) {
      // QUANT path.

For now we just have this global repository of ADRs.
If we see fit, we can break these out to per-platform ones, or keep a hybrid approach to having global and platform-specific ADSs.

Our aim is to keep the process lightweight and approachable.

// RUN: stablehlo-quant-opt %s -split-input-file -stablehlo-quantize | FileCheck %s

// Test that hybrid quantized dot_general is produced when q/dq pair only exists
// for weight.

module attributes {tf_saved_model.semantics} {
  func.func private @quantize_dot_general_fn(%arg0: tensor<1x2xf32>) -> tensor<1x3xf32> attributes {tf._original_func_name = "main_0"} {
    %cst = stablehlo.constant dense<3.000000e-01> : tensor<2x3xf32>

		t.Fatalf("verification failed for google.com chain (system only pool): %s", err)
	}

	pool.AddCert(root)

	_, err = googChain[0].Verify(opts)
	if err != nil {
		t.Fatalf("verification failed for google.com chain (hybrid pool): %s", err)
	}

	certTmpl := &x509.Certificate{
		SerialNumber: big.NewInt(1),
		NotBefore:    time.Now().Add(-time.Hour),
		NotAfter:     time.Now().Add(time.Hour * 10),

  }

  // We should not observe both uint8 & int8.
  if (int8_type_observed && uint8_type_observed) return UNKNOWN;

  if (float_type_observed) {
    if (int8_type_observed || uint8_type_observed) {
      return HYBRID;
    } else {
      return FLOAT;
    }
  }

  if (int8_type_observed) {
    return QUANTIZED_INT8;
  }

  if (uint8_type_observed) {
    return QUANTIZED_UINT8;
  }

// point constant.
def : Pat<(TFL_DequantizeOp
             (TFL_QuantizeOp (Arith_ConstantOp F32ElementsAttr:$cst), $qt)),
          (TFL_ConstOp $cst)>;

// Transpose conv supports hybrid computation with quantized weights.
def FoldQuantWeightsIntoTposeConv : Pat<
  (TFL_TransposeConvOp
    $output_shape,
    (TFL_DequantizeOp $quant_weights),
    $quant_input,