💡 All-in-one AI framework for semantic search, LLM orchestration and language model workflows
Python agentsaiai-agentsembeddingsinformation-retrievallanguage-modellarge-language-modelsllmnlppythonragretrieval-augmented-generationsearchsearch-enginesemantic-searchsentence-embeddingstransformerstxtaivector-databasevector-search
"""
Zvec ANN module tests
"""
import numpy as np
from txtai.ann import ANNFactory
from .base import DenseTest
class TestZvec(DenseTest):
"""
Zvec ANN tests.
"""
def testZvec(self):
"""
Test zvec backend
"""
self.runTests("zvec")
self.deletenumpy("zvec")
def testZvecCustom(self):
"""
Test zvec backend with custom settings
"""
self.runTests("zvec", {"zvec": {"efconstruction": 100, "m": 16, "efsearch": 50}})
# Test invalid file path handled
with self.assertRaises(FileNotFoundError):
ann = ANNFactory.create({"backend": "zvec"})
ann.load("non-exist-path")
def testZvecDatatype(self):
"""
Test zvec backend vector data type setting
"""
# fp32 is the default, fp16 is a supported alternative
for datatype, expected in (("fp32", "VECTOR_FP32"), ("fp16", "VECTOR_FP16")):
self.runTests("zvec", {"zvec": {"datatype": datatype}}, update=False)
ann = ANNFactory.create({"backend": "zvec", "dimensions": 8, "zvec": {"datatype": datatype}})
self.assertEqual(ann.datatype().name, expected)
# int8 stores integer vectors
data = np.random.randint(0, 100, (100, 8)).astype(np.int8)
ann = ANNFactory.create({"backend": "zvec", "dimensions": data.shape[1], "zvec": {"datatype": "int8"}})
ann.index(data)
self.assertEqual(ann.count(), 100)
self.assertEqual(len(ann.search(data[:1], 1)[0]), 1)
# Unsupported values are rejected with the supported options
ann = ANNFactory.create({"backend": "zvec", "dimensions": 8, "zvec": {"datatype": "fp64"}})
with self.assertRaises(ValueError):
ann.index(np.random.rand(4, 8).astype(np.float32))