Add GPT4Vision Image loader (#1089)

Co-authored-by: Deshraj Yadav <deshrajdry@gmail.com>
This commit is contained in:
Sidharth Mohanty
2024-01-02 03:57:23 +05:30
committed by GitHub
parent 367d6b70e2
commit c62663f2e4
29 changed files with 291 additions and 714 deletions

View File

@@ -1,78 +0,0 @@
import unittest
from embedchain.chunkers.images import ImagesChunker
from embedchain.config import ChunkerConfig
from embedchain.models.data_type import DataType
class TestImageChunker(unittest.TestCase):
def test_chunks(self):
"""
Test the chunks generated by TextChunker.
# TODO: Not a very precise test.
"""
chunker_config = ChunkerConfig(chunk_size=1, chunk_overlap=0, length_function=len, min_chunk_size=0)
chunker = ImagesChunker(config=chunker_config)
# Data type must be set manually in the test
chunker.set_data_type(DataType.IMAGES)
image_path = "./tmp/image.jpeg"
app_id = "app1"
result = chunker.create_chunks(MockLoader(), image_path, app_id=app_id)
expected_chunks = {
"doc_id": f"{app_id}--123",
"documents": [image_path],
"embeddings": ["embedding"],
"ids": ["140bedbf9c3f6d56a9846d2ba7088798683f4da0c248231336e6a05679e4fdfe"],
"metadatas": [{"data_type": "images", "doc_id": f"{app_id}--123", "url": "none"}],
}
self.assertEqual(expected_chunks, result)
def test_chunks_with_default_config(self):
"""
Test the chunks generated by ImageChunker with default config.
"""
chunker = ImagesChunker()
# Data type must be set manually in the test
chunker.set_data_type(DataType.IMAGES)
image_path = "./tmp/image.jpeg"
app_id = "app1"
result = chunker.create_chunks(MockLoader(), image_path, app_id=app_id)
expected_chunks = {
"doc_id": f"{app_id}--123",
"documents": [image_path],
"embeddings": ["embedding"],
"ids": ["140bedbf9c3f6d56a9846d2ba7088798683f4da0c248231336e6a05679e4fdfe"],
"metadatas": [{"data_type": "images", "doc_id": f"{app_id}--123", "url": "none"}],
}
self.assertEqual(expected_chunks, result)
def test_word_count(self):
chunker_config = ChunkerConfig(chunk_size=1, chunk_overlap=0, length_function=len, min_chunk_size=0)
chunker = ImagesChunker(config=chunker_config)
chunker.set_data_type(DataType.IMAGES)
document = [["ab cd", "ef gh"], ["ij kl", "mn op"]]
result = chunker.get_word_count(document)
self.assertEqual(result, 1)
class MockLoader:
def load_data(self, src):
"""
Mock loader that returns a list of data dictionaries.
Adjust this method to return different data for testing.
"""
return {
"doc_id": "123",
"data": [
{
"content": src,
"embedding": "embedding",
"meta_data": {"url": "none"},
}
],
}