test: use ml-dtypes lib to produce bf16 datatype (#33354)

Signed-off-by: zhuwenxing <wenxing.zhu@zilliz.com>

tests/python_client/common/bulk_insert_data.py

@@ -4,7 +4,7 @@ import os
 import time
 
 import numpy as np
-import jax.numpy as jnp
+from ml_dtypes import bfloat16
 import pandas as pd
 import random
 from faker import Faker
@@ -128,9 +128,9 @@ def gen_bf16_vectors(num, dim, for_json=False):
         raw_vector = [random.random() for _ in range(dim)]
         raw_vectors.append(raw_vector)
         if for_json:
-            bf16_vector = np.array(jnp.array(raw_vector, dtype=jnp.bfloat16)).tolist()
+            bf16_vector = np.array(raw_vector, dtype=bfloat16).tolist()
         else:
-            bf16_vector = np.array(jnp.array(raw_vector, dtype=jnp.bfloat16)).view(np.uint8).tolist()
+            bf16_vector = np.array(raw_vector, dtype=bfloat16).view(np.uint8).tolist()
         bf16_vectors.append(bf16_vector)
 
     return raw_vectors, bf16_vectors

tests/python_client/common/common_func.py

@@ -8,7 +8,7 @@ import uuid
 from functools import singledispatch
 import numpy as np
 import pandas as pd
-import jax.numpy as jnp
+from ml_dtypes import bfloat16
 from sklearn import preprocessing
 from npy_append_array import NpyAppendArray
 from faker import Faker
@@ -20,7 +20,6 @@ from common import common_type as ct
 from utils.util_log import test_log as log
 from customize.milvus_operator import MilvusOperator
 import pickle
-import tensorflow as tf
 fake = Faker()
 """" Methods of processing data """
 
@@ -1070,14 +1069,12 @@ def gen_data_by_collection_field(field, nb=None, start=None):
         dim = field.params['dim']
         if nb is None:
             raw_vector = [random.random() for _ in range(dim)]
-            bf16_vector = jnp.array(raw_vector, dtype=jnp.bfloat16)
-            bf16_vector = np.array(bf16_vector).view(np.uint8).tolist()
+            bf16_vector = np.array(raw_vector, dtype=bfloat16).view(np.uint8).tolist()
             return bytes(bf16_vector)
         bf16_vectors = []
         for i in range(nb):
             raw_vector = [random.random() for _ in range(dim)]
-            bf16_vector = jnp.array(raw_vector, dtype=jnp.bfloat16)
-            bf16_vector = np.array(bf16_vector).view(np.uint8).tolist()
+            bf16_vector = np.array(raw_vector, dtype=bfloat16).view(np.uint8).tolist()
             bf16_vectors.append(bytes(bf16_vector))
         return bf16_vectors
     if data_type == DataType.FLOAT16_VECTOR:
@@ -2077,7 +2074,7 @@ def gen_bf16_vectors(num, dim):
     for _ in range(num):
         raw_vector = [random.random() for _ in range(dim)]
         raw_vectors.append(raw_vector)
-        bf16_vector = tf.cast(raw_vector, dtype=tf.bfloat16).numpy()
+        bf16_vector = np.array(raw_vector, dtype=bfloat16)
         bf16_vectors.append(bf16_vector)
 
     return raw_vectors, bf16_vectors

tests/python_client/requirements.txt

@@ -56,7 +56,5 @@ pyarrow==14.0.1
 fastparquet==2023.7.0
 
 # for bf16 datatype
-jax==0.4.13
-jaxlib==0.4.13
-tensorflow==2.13.1
+ml-dtypes==0.2.0