fix test

Author: metascroy

kernels/quantized/test/op_embedding2b_test.cpp

@@ -36,19 +36,19 @@ TEST(OpQuantizedEmbedding2bTest, TestGroupWiseQuantizedEmbedding) {
   int64_t quant_max = 1;
 
   Tensor weight_scales = tf.make({3}, {0.5, 1.0, 1.5});
-  Tensor weight_zero_points = tf.make({3}, {1, -5, 0});
+  Tensor weight_zero_points = tf.make({3}, {1, -2, 0});
 
-  // -3,  1,  6, 7,
-  //  2, -5, -4, 0,
-  // -8,  3, -1, 6,
+  // -2,  1,  0, 1, -> 0, 3, 2, 3 -> 00 11 10 11 -> 59
+  //  0, -1, -2, 0, -> 2, 1, 0, 2 -> 10 01 00 10 -> 146
+  // -2,  -1, 0, 1, -> 0, 1, 2, 3 -> 00 01 10 11 -> 27
 
-  Tensor qweight = tfb.make({3, 2}, {89, 239, 163, 72, 11, 126});
+  Tensor qweight = tfb.make({3, 1}, {59, 146, 27});
 
   Tensor indices = tfl.make({3}, {0, 2, 1});
 
   Tensor out = tf.zeros({3, 4});
-  Tensor expected = tf.make(
-      {3, 4}, {-2.0, 0.0, 2.5, 3.0, -12.0, 4.5, -1.5, 9.0, 7.0, 0.0, 1.0, 5.0});
+  Tensor expected = tf.make( 
+      {3, 4}, {-1.5, 0.0, -0.5, 0.0, -3.0, -1.5, 0.0, 1.5, -2.0, -3.0, -4.0, -2.0});
 
   quantized_embedding_2bit_out(
       qweight,
@@ -76,18 +76,21 @@ TEST(OpQuantizedEmbedding2bTest, TestGroupWiseQuantizedEmbedding) {
   EXPECT_TENSOR_EQ(out, expected);
 
   // Groupwise quantization. groupsize = 2
+  
   weight_scales = tf.make({3, 2}, {0.5, 1.0, 1.5, 2.0, 2.5, 3.0});
-  weight_zero_points = tf.make({3, 2}, {1, -5, 0, 2, -3, -1});
-  /*
-  fp_weight = [-2.0,  0.0,  11.0, 12.0,
-                3.0, -7.5, -12.0, -4.0,
-              -12.5, 15.0,   0.0, 21.0]
-  */
+  weight_zero_points = tf.make({3, 2}, {1, -2, 0, 1, -2, -1});
 
-  out = tf.zeros({3, 4});
-  expected = tf.make(
-      {3, 4},
-      {-2.0, 0.0, 11.0, 12.0, -12.5, 15.0, 0.0, 21.0, 3.0, -7.5, -12.0, -4.0});
+  // -2,  1,  0, 1, -> 0, 3, 2, 3 -> 00 11 10 11 -> 59
+  //  0, -1, -2, 0, -> 2, 1, 0, 2 -> 10 01 00 10 -> 146
+  // -2,  -1, 0, 1, -> 0, 1, 2, 3 -> 00 01 10 11 -> 27
+
+  Tensor qweight = tfb.make({3, 1}, {59, 146, 27});
+
+  Tensor indices = tfl.make({3}, {0, 2, 1});
+
+  Tensor out = tf.zeros({3, 4});
+  Tensor expected = tf.make( 
+      {3, 4}, {-1.5, 0.0, -2.0, -1.0, 0.0, 2.5, 3.0, 6.0, 0.0, -1.5, -6.0, -2.0});
 
   quantized_embedding_2bit_out(
       qweight,
@@ -111,11 +114,11 @@ TEST(OpQuantizedEmbedding2bTest, TestGroupWiseQuantizedEmbeddingDeath1) {
   int64_t quant_max = 1;
 
   Tensor weight_scales = tf.make({4}, {0.5, 1.0, 1.5, 3.3});
-  Tensor weight_zero_points = tf.make({4}, {1, 5, 7, 5});
-  Tensor qweight = tfb.make({3, 2}, {89, 239, 163, 72, 11, 126});
+  Tensor weight_zero_points = tf.make({4}, {1, -2, 1, 0});
+  Tensor qweight = tfb.make({3, 1}, {59, 146, 27});
   Tensor indices = tfl.make({3}, {0, 2, 1});
-
   Tensor out = tf.zeros({3, 4});
+
   ET_EXPECT_DEATH(
       quantized_embedding_2bit_out(
           qweight,
@@ -138,9 +141,10 @@ TEST(OpQuantizedEmbedding2bTest, TestGroupWiseQuantizedEmbeddingDeath2) {
   int64_t quant_max = 1;
 
   Tensor weight_scales = tf.make({2}, {0.5, 1.0});
-  Tensor weight_zero_points = tf.make({2}, {1, 5});
-  Tensor qweight = tfb.make({3, 2}, {89, 239, 163, 72, 11, 126});
+  Tensor weight_zero_points = tf.make({2}, {1, -2});
+  Tensor qweight = tfb.make({3, 1}, {59, 146, 27});
   Tensor indices = tfl.make({3}, {0, 2, 1});
+  Tensor out = tf.zeros({3, 4});
 
   Tensor out = tf.zeros({3, 4});
   ET_EXPECT_DEATH(