v0.12.25

Compare Source

• Ensuring original text is preserved in CHUNKING_REGEX in splitters (#​18054)
• use SimpleDirectoryReader without llama-index-readers-file package (#​18173)
• Improved Annotations and Error Handling in utils.py and exec_utils.py (#​18153)
• Add build_semantic_nodes_from_nodes to SemanticDoubleMergingSplitterNodeParser (#​18114)
• fix: Optimize memory management of the Context object (#​18170)
• fix: allow streaming events from context after workflows ends (#​18174)
• feat: Add a clear method to the Context class (#​18136)

v0.12.24.post1

Compare Source

v0.12.24

• fix get content node metadata duplicated templates (#​18110)
• remove assert for multimodal llms in multimodal vector store/query engine (#​18112)
• fix: make run_step return all the events produced by a step (#​18082)
• fix(multimodal nodes): check image path is file before open (#​18043)

v0.4.0

Compare Source

• Added top-level pgvector package
• Added support for pg8000
• Added support for bytes to Bit constructor
• Changed globally option to default to False for Psycopg 2
• Changed arrays option to default to True for Psycopg 2
• Fixed equality for Vector, HalfVector, Bit, and SparseVector classes
• Fixed indices and values methods of SparseVector returning tuple instead of list in some cases
• Dropped support for Python < 3.9