[Bug]: Using GPU_IVF_FLAT and COSINE wtih the same parameters as IVF_FLAT brings errors

[qwevdb]

Is there an existing issue for this?

• I have searched the existing issues

Environment

- Milvus version: milvus v2.4.12-gpu
- Deployment mode(standalone or cluster): standalone
- MQ type(rocksmq, pulsar or kafka): rocksmq   
- SDK version(e.g. pymilvus v2.0.0rc2): pymilvus v2.4.5
- OS(Ubuntu or CentOS): Ubuntu 24.04 LTS
- CPU/Memory: Intel Core i7-11700 / 64G
- GPU: NVIDIA GeForce RTX 4090
- Others:

Current Behavior

Using the IVF_FLAT index and COSINE metric can normally output results, while using the GPU_IVF_FLAT index and COSINE metric with the same parameters produce errors.

Expected Behavior

Both IVF_FLAT and GPU_IVF_FLAT index with COSINE metric and the same parameters can produce the same results.

Steps To Reproduce

1. run python script as follow

import time
from pymilvus import Collection, connections, FieldSchema, CollectionSchema, DataType, utility
import numpy as np

FLOAT_MAX = 5000
DATA_INT_MAX = 100

numpy_random = np.random.default_rng(0)
alias = "bench"
collection_name = "Benchmark"
client = connections.connect(
   alias=alias,
   host="localhost",
   port="19530"
)
if utility.has_collection(collection_name, using=alias):
   collection = Collection(name=collection_name, using=alias)
   collection.drop()
   time.sleep(2)  
   
dim = 275
id = FieldSchema(name='id', dtype=DataType.INT64, is_primary=True)
vector = FieldSchema(name='vector', dtype=DataType.FLOAT_VECTOR, dim=dim)
fields = [id, vector]
schema = CollectionSchema(fields=fields, description=alias)
collection = Collection(
   name=collection_name,
   schema=schema,
   using=alias,
)
index_params = {'index_type': 'IVF_FLAT', 'params': {'nlist': 57953}, 'metric_type': 'COSINE'}
# index_params = {'index_type': 'GPU_IVF_FLAT', 'params': {'nlist': 57953}, 'metric_type': 'COSINE'}
collection.create_index("vector", index_params, timeout=100)

dataset = []
number = 2297
for i in range(0, number + 0):
   vector = numpy_random.random((1, dim))
   data = {
       'id': i,
       'vector': list(vector[0]),

   }
   dataset.append(data)
collection.upsert(dataset)
collection.flush()
collection.load()

dataset = []
number = 64
for i in range(2297,number + 2297):
   vector = numpy_random.random((1, dim))
   data = {
       'id': i,
       'vector': list(vector[0]),

   }
   dataset.append(data)
collection.insert(dataset)
collection.flush()
collection.load()

vector = numpy_random.random((1, dim))
query_vector = list(vector[0])
iterator = collection.search_iterator(
      data=[query_vector],
      anns_field="vector",
      param={"metric_type": "COSINE", "params": {'nprobe': 13289}},
      limit=8,
      expr=None,
      batch_size=6609,
      timeout=100
   )
res1 = []
while True:
   result = iterator.next()
   if not result:
       iterator.close()
       break

   res1.extend(result)
collection.release()
collection.drop_index()
collection.flush()
print(res1)
collection.drop()

result:

[id: 1303, distance: 0.7831529378890991, entity: {}, id: 2142, distance: 0.782934844493866, entity: {}, id: 893, distance: 0.7822405099868774, entity: {}, id: 1443, distance: 0.7811518907546997, entity: {}, id: 1333, distance: 0.7790765762329102, entity: {}, id: 2308, distance: 0.7782455682754517, entity: {}, id: 32, distance: 0.7777007818222046, entity: {}, id: 490, distance: 0.777592658996582, entity: {}]

2. change 'index_type': 'IVF_FLAT' to 'index_type': 'GPU_IVF_FLAT' in index_params and run again

result:

RPC error: [create_index], <MilvusException: (code=1100, message=metric type not found or not supported, supported: [L2 IP]: invalid parameter[expected=valid index params][actual=invalid index params])>, <Time:{'RPC start': '2024-09-29 22:15:23.000733', 'RPC error': '2024-09-29 22:15:23.002850'}>
Traceback (most recent call last):
  File "/home/xxx/Desktop/vdb/milvus/vdb-test/./log/milvus/test_bug1_B.py", line 32, in <module>
    collection.create_index("vector", index_params1, timeout=100)
  File "/root/anaconda3/envs/milvus/lib/python3.10/site-packages/pymilvus/orm/collection.py", line 1373, in create_index
    return conn.create_index(self._name, field_name, index_params, timeout=timeout, **kwargs)
  File "/root/anaconda3/envs/milvus/lib/python3.10/site-packages/pymilvus/decorators.py", line 148, in handler
    raise e from e
  File "/root/anaconda3/envs/milvus/lib/python3.10/site-packages/pymilvus/decorators.py", line 144, in handler
    return func(*args, **kwargs)
  File "/root/anaconda3/envs/milvus/lib/python3.10/site-packages/pymilvus/decorators.py", line 183, in handler
    return func(self, *args, **kwargs)
  File "/root/anaconda3/envs/milvus/lib/python3.10/site-packages/pymilvus/decorators.py", line 123, in handler
    raise e from e
  File "/root/anaconda3/envs/milvus/lib/python3.10/site-packages/pymilvus/decorators.py", line 87, in handler
    return func(*args, **kwargs)
  File "/root/anaconda3/envs/milvus/lib/python3.10/site-packages/pymilvus/client/grpc_handler.py", line 987, in create_index
    check_status(status)
  File "/root/anaconda3/envs/milvus/lib/python3.10/site-packages/pymilvus/client/utils.py", line 63, in check_status
    raise MilvusException(status.code, status.reason, status.error_code)
pymilvus.exceptions.MilvusException: <MilvusException: (code=1100, message=metric type not found or not supported, supported: [L2 IP]: invalid parameter[expected=valid index params][actual=invalid index params])>

[xiaofan-luan]

GPU index only support L2 IP, cosine is not supported yet.
However, you can normalize vector so IP distance is same as L2

[yanliang567]

/assign @qwevdb
/unassign

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.
Rotten issues close after 30d of inactivity. Reopen the issue with /reopen.