| from ragflow_sdk import RAGFlow | |
| from common import HOST_ADDRESS | |
| from time import sleep | |
| def test_parse_document_with_txt(get_api_key_fixture): | |
| API_KEY = get_api_key_fixture | |
| rag = RAGFlow(API_KEY, HOST_ADDRESS) | |
| ds = rag.create_dataset(name="test_parse_document") | |
| name = 'ragflow_test.txt' | |
| with open("test_data/ragflow_test.txt","rb") as file : | |
| blob = file.read() | |
| docs = ds.upload_documents([{"displayed_name": name, "blob": blob}]) | |
| doc = docs[0] | |
| ds.async_parse_documents(document_ids=[doc.id]) | |
| ''' | |
| for n in range(100): | |
| if doc.progress == 1: | |
| break | |
| sleep(1) | |
| else: | |
| raise Exception("Run time ERROR: Document parsing did not complete in time.") | |
| ''' | |
| def test_parse_and_cancel_document(get_api_key_fixture): | |
| API_KEY = get_api_key_fixture | |
| rag = RAGFlow(API_KEY, HOST_ADDRESS) | |
| ds = rag.create_dataset(name="test_parse_and_cancel_document") | |
| name = 'ragflow_test.txt' | |
| with open("test_data/ragflow_test.txt","rb") as file : | |
| blob = file.read() | |
| docs=ds.upload_documents([{"displayed_name": name, "blob": blob}]) | |
| doc = docs[0] | |
| ds.async_parse_documents(document_ids=[doc.id]) | |
| sleep(1) | |
| if 0 < doc.progress < 1: | |
| ds.async_cancel_parse_documents(document_ids=[doc.id]) | |
| def test_bulk_parse_documents(get_api_key_fixture): | |
| API_KEY = get_api_key_fixture | |
| rag = RAGFlow(API_KEY, HOST_ADDRESS) | |
| ds = rag.create_dataset(name="test_bulk_parse_and_cancel_documents") | |
| with open("ragflow.txt","rb") as file: | |
| blob = file.read() | |
| documents = [ | |
| {'displayed_name': 'test1.txt', 'blob': blob}, | |
| {'displayed_name': 'test2.txt', 'blob': blob}, | |
| {'displayed_name': 'test3.txt', 'blob': blob} | |
| ] | |
| docs = ds.upload_documents(documents) | |
| ids = [doc.id for doc in docs] | |
| ds.async_parse_documents(ids) | |
| ''' | |
| for n in range(100): | |
| all_completed = all(doc.progress == 1 for doc in docs) | |
| if all_completed: | |
| break | |
| sleep(1) | |
| else: | |
| raise Exception("Run time ERROR: Bulk document parsing did not complete in time.") | |
| ''' | |
| def test_list_chunks_with_success(get_api_key_fixture): | |
| API_KEY = get_api_key_fixture | |
| rag = RAGFlow(API_KEY, HOST_ADDRESS) | |
| ds = rag.create_dataset(name="test_list_chunks_with_success") | |
| with open("test_data/ragflow_test.txt", "rb") as file: | |
| blob = file.read() | |
| ''' | |
| # chunk_size = 1024 * 1024 | |
| # chunks = [blob[i:i + chunk_size] for i in range(0, len(blob), chunk_size)] | |
| documents = [ | |
| {'displayed_name': f'chunk_{i}.txt', 'blob': chunk} for i, chunk in enumerate(chunks) | |
| ] | |
| ''' | |
| documents =[{"displayed_name":"test_list_chunks_with_success.txt","blob":blob}] | |
| docs = ds.upload_documents(documents) | |
| ids = [doc.id for doc in docs] | |
| ds.async_parse_documents(ids) | |
| ''' | |
| for n in range(100): | |
| all_completed = all(doc.progress == 1 for doc in docs) | |
| if all_completed: | |
| break | |
| sleep(1) | |
| else: | |
| raise Exception("Run time ERROR: Chunk document parsing did not complete in time.") | |
| ''' | |
| doc = docs[0] | |
| doc.list_chunks() | |
| def test_add_chunk_with_success(get_api_key_fixture): | |
| API_KEY = get_api_key_fixture | |
| rag = RAGFlow(API_KEY, HOST_ADDRESS) | |
| ds = rag.create_dataset(name="test_add_chunk_with_success") | |
| with open("test_data/ragflow_test.txt", "rb") as file: | |
| blob = file.read() | |
| ''' | |
| # chunk_size = 1024 * 1024 | |
| # chunks = [blob[i:i + chunk_size] for i in range(0, len(blob), chunk_size)] | |
| documents = [ | |
| {'displayed_name': f'chunk_{i}.txt', 'blob': chunk} for i, chunk in enumerate(chunks) | |
| ] | |
| ''' | |
| documents =[{"displayed_name":"test_list_chunks_with_success.txt","blob":blob}] | |
| docs = ds.upload_documents(documents) | |
| doc = docs[0] | |
| doc.add_chunk(content="This is a chunk addition test") | |
| def test_delete_chunk_with_success(get_api_key_fixture): | |
| API_KEY = get_api_key_fixture | |
| rag = RAGFlow(API_KEY, HOST_ADDRESS) | |
| ds = rag.create_dataset(name="test_delete_chunk_with_success") | |
| with open("test_data/ragflow_test.txt", "rb") as file: | |
| blob = file.read() | |
| ''' | |
| # chunk_size = 1024 * 1024 | |
| # chunks = [blob[i:i + chunk_size] for i in range(0, len(blob), chunk_size)] | |
| documents = [ | |
| {'displayed_name': f'chunk_{i}.txt', 'blob': chunk} for i, chunk in enumerate(chunks) | |
| ] | |
| ''' | |
| documents =[{"displayed_name":"test_delete_chunk_with_success.txt","blob":blob}] | |
| docs = ds.upload_documents(documents) | |
| doc = docs[0] | |
| chunk = doc.add_chunk(content="This is a chunk addition test") | |
| doc.delete_chunks([chunk.id]) | |
| def test_update_chunk_content(get_api_key_fixture): | |
| API_KEY = get_api_key_fixture | |
| rag = RAGFlow(API_KEY, HOST_ADDRESS) | |
| ds = rag.create_dataset(name="test_update_chunk_content_with_success") | |
| with open("test_data/ragflow_test.txt", "rb") as file: | |
| blob = file.read() | |
| ''' | |
| # chunk_size = 1024 * 1024 | |
| # chunks = [blob[i:i + chunk_size] for i in range(0, len(blob), chunk_size)] | |
| documents = [ | |
| {'displayed_name': f'chunk_{i}.txt', 'blob': chunk} for i, chunk in enumerate(chunks) | |
| ] | |
| ''' | |
| documents =[{"displayed_name":"test_update_chunk_content_with_success.txt","blob":blob}] | |
| docs = ds.upload_documents(documents) | |
| doc = docs[0] | |
| chunk = doc.add_chunk(content="This is a chunk addition test") | |
| chunk.update({"content":"This is a updated content"}) | |
| def test_update_chunk_available(get_api_key_fixture): | |
| API_KEY = get_api_key_fixture | |
| rag = RAGFlow(API_KEY, HOST_ADDRESS) | |
| ds = rag.create_dataset(name="test_update_chunk_available_with_success") | |
| with open("test_data/ragflow_test.txt", "rb") as file: | |
| blob = file.read() | |
| ''' | |
| # chunk_size = 1024 * 1024 | |
| # chunks = [blob[i:i + chunk_size] for i in range(0, len(blob), chunk_size)] | |
| documents = [ | |
| {'displayed_name': f'chunk_{i}.txt', 'blob': chunk} for i, chunk in enumerate(chunks) | |
| ] | |
| ''' | |
| documents =[{"displayed_name":"test_update_chunk_available_with_success.txt","blob":blob}] | |
| docs = ds.upload_documents(documents) | |
| doc = docs[0] | |
| chunk = doc.add_chunk(content="This is a chunk addition test") | |
| chunk.update({"available":False}) | |
| def test_retrieve_chunks(get_api_key_fixture): | |
| API_KEY = get_api_key_fixture | |
| rag = RAGFlow(API_KEY, HOST_ADDRESS) | |
| ds = rag.create_dataset(name="retrieval") | |
| with open("test_data/ragflow_test.txt", "rb") as file: | |
| blob = file.read() | |
| ''' | |
| # chunk_size = 1024 * 1024 | |
| # chunks = [blob[i:i + chunk_size] for i in range(0, len(blob), chunk_size)] | |
| documents = [ | |
| {'displayed_name': f'chunk_{i}.txt', 'blob': chunk} for i, chunk in enumerate(chunks) | |
| ] | |
| ''' | |
| documents =[{"displayed_name":"test_retrieve_chunks.txt","blob":blob}] | |
| docs = ds.upload_documents(documents) | |
| doc = docs[0] | |
| doc.add_chunk(content="This is a chunk addition test") | |
| rag.retrieve(dataset_ids=[ds.id],document_ids=[doc.id]) |