enhance: Multi-cp from master to 2.5.x (#2668)

- fix: type annotation in Hits and docs in Collection (#2387)
- enhance: update float16/bfloat16 examples (#2388)
- fix: sparse slice bug in hello_model.py(#2414) (#2417)
- fix: pkg_resources api deprecated warning (#2438)
- enhance: Use new model pkg (#2595)

See also: milvus-io/milvus#37448

Signed-off-by: yangxuan <[email protected]>
Co-authored-by: Yinzuo Jiang <[email protected]>
Co-authored-by: Yinzuo Jiang <[email protected]>
Co-authored-by: jinjuan zhou <[email protected]>
Co-authored-by: zhuwenxing <[email protected]>
Co-authored-by: junjie.jiang <[email protected]>

Author: 5 people

examples/datatypes/bfloat16_example.py

@@ -2,6 +2,7 @@
 import random
 import numpy as np
 import tensorflow as tf
+import torch
 from pymilvus import (
      connections,
      utility,
@@ -20,6 +21,11 @@ def gen_bf16_vectors(num, dim):
     for _ in range(num):
         raw_vector = [random.random() for _ in range(dim)]
         raw_vectors.append(raw_vector)
+        # Numpy itself does not support bfloat16, use TensorFlow extension instead.
+        # PyTorch does not support converting bfloat16 vector to numpy array.
+        # See:
+        # - https://github.com/numpy/numpy/issues/19808
+        # - https://github.com/pytorch/pytorch/issues/90574
         bf16_vector = tf.cast(raw_vector, dtype=tf.bfloat16).numpy()
         bf16_vectors.append(bf16_vector)
     return raw_vectors, bf16_vectors
@@ -57,8 +63,10 @@ def bf16_vector_search():
                                   index_params={"index_type": index_type, "params": index_params, "metric_type": "L2"})
         hello_milvus.load()
         print("index_type = ", index_type)
-        res = hello_milvus.search(vectors[0:10], vector_field_name, {"metric_type": "L2"}, limit=1)
-        print(res)
+        res = hello_milvus.search(vectors[0:10], vector_field_name, {"metric_type": "L2"}, limit=1, output_fields=["bfloat16_vector"])
+        print("raw bytes: ", res[0][0].get("bfloat16_vector"))
+        print("tensorflow Tensor: ", tf.io.decode_raw(res[0][0].get("bfloat16_vector"), tf.bfloat16, little_endian=True))
+        print("pytorch Tensor: ", torch.frombuffer(res[0][0].get("bfloat16_vector"), dtype=torch.bfloat16))
         hello_milvus.release()
         hello_milvus.drop_index()
 

examples/datatypes/float16_example.py

@@ -13,13 +13,16 @@
 
 default_fp16_index_params = [{"nlist": 128}]
 
+# float16, little endian
+fp16_little = np.dtype('e').newbyteorder('<')
+
 def gen_fp16_vectors(num, dim):
     raw_vectors = []
     fp16_vectors = []
     for _ in range(num):
         raw_vector = [random.random() for _ in range(dim)]
         raw_vectors.append(raw_vector)
-        fp16_vector = np.array(raw_vector, dtype=np.float16)
+        fp16_vector = np.array(raw_vector, dtype=fp16_little)
         fp16_vectors.append(fp16_vector)
     return raw_vectors, fp16_vectors
 
@@ -57,8 +60,9 @@ def fp16_vector_search():
                                   index_params={"index_type": index_type, "params": index_params, "metric_type": "L2"})
         hello_milvus.load()
         print("index_type = ", index_type)
-        res = hello_milvus.search(vectors[0:10], vector_field_name, {"metric_type": "L2"}, limit=1)
-        print(res)
+        res = hello_milvus.search(vectors[0:10], vector_field_name, {"metric_type": "L2"}, limit=1, output_fields=["float16_vector"])
+        print("raw bytes: ", res[0][0].get("float16_vector"))
+        print("numpy ndarray: ", np.frombuffer(res[0][0].get("float16_vector"), dtype=fp16_little))
         hello_milvus.release()
         hello_milvus.drop_index()
 

examples/milvus_model/hello_model.py

@@ -52,21 +52,16 @@ def log(msg):
 # BM25EmbeddingFunction usage
 log(fmt.format("BM25EmbeddingFunction Usage"))
 ef_bm25 = BM25EmbeddingFunction()
-docs_bm25 = [
-    "Artificial intelligence was founded as an academic discipline in 1956.",
-    "Alan Turing was the first person to conduct substantial research in AI.",
-    "Born in Maida Vale, London, Turing was raised in southern England.",
-]
 ef_bm25.load()
 embs_bm25 = ef_bm25.encode_documents(docs)
-log(f"Embedding Shape: {embs_bm25[0].shape} Dimension: {ef_bm25.dim}")
+log(f"Embedding Shape: {embs_bm25[:, [0]].shape} Dimension: {ef_bm25.dim}")
 
 # -----------------------------------------------------------------------------
 # SpladeEmbeddingFunction usage
 log(fmt.format("SpladeEmbeddingFunction Usage"))
 ef_splade = SpladeEmbeddingFunction(device="cpu")
 embs_splade = ef_splade(["Hello world", "Hello world2"])
-log(f"Embedding Shape: {embs_splade[0].shape} Dimension: {ef_splade.dim}")
+log(f"Embedding Shape: {embs_splade[:, [0]].shape} Dimension: {ef_splade.dim}")
 
 # -----------------------------------------------------------------------------
 log(fmt.format("Demonstrations Finished"))

pymilvus/client/__init__.py

@@ -2,17 +2,16 @@
 import re
 import subprocess
 from contextlib import suppress
-
-from pkg_resources import DistributionNotFound, get_distribution
+from importlib.metadata import PackageNotFoundError, version
 
 log = logging.getLogger(__name__)
 
 
 __version__ = "0.0.0.dev"
 
 
-with suppress(DistributionNotFound):
-    __version__ = get_distribution("pymilvus").version
+with suppress(PackageNotFoundError):
+    __version__ = version("pymilvus")
 
 
 def get_commit(version: str = "", short: bool = True) -> str:

pymilvus/client/abstract.py

@@ -679,7 +679,7 @@ class Hits(list):
     def __init__(
         self,
         topk: int,
-        pks: Union[int, str],
+        pks: List[Union[int, str]],
         distances: List[float],
         fields: Dict[str, Tuple[List[Any], schema_pb2.FieldData]],
         output_fields: List[str],

pymilvus/model/__init__.py

@@ -698,7 +698,7 @@ def search(
 
             partition_names (``List[str]``, optional): The names of partitions to search on.
             output_fields (``List[str]``, optional):
-                The name of fields to return in the search result.  Can only get scalar fields.
+                The name of fields to return in the search result.
             round_decimal (``int``, optional):
                 The specified number of decimal places of returned distance.
                 Defaults to -1 means no round to returned distance.
@@ -833,7 +833,7 @@ def hybrid_search(
 
             partition_names (``List[str]``, optional): The names of partitions to search on.
             output_fields (``List[str]``, optional):
-                The name of fields to return in the search result.  Can only get scalar fields.
+                The name of fields to return in the search result.
             round_decimal (``int``, optional):
                 The specified number of decimal places of returned distance.
                 Defaults to -1 means no round to returned distance.

pymilvus/orm/collection.py

@@ -52,7 +52,7 @@ bulk_writer = [
 ]
 
 model = [
-    "milvus-model>=0.1.0",
+    "pymilvus.model>=0.3.0",
 ]
 
 dev = [