From 536d6f2283815fedb2c423010504fb12fc440055 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?=E7=AB=A5=E7=9F=B3=E6=B8=8A?= Date: Tue, 7 Jan 2025 00:28:15 +0800 Subject: [PATCH 1/3] =?UTF-8?q?=E6=B7=BB=E5=8A=A0=E5=AD=97=E7=AC=A6?= =?UTF-8?q?=E5=88=86=E5=89=B2=E5=8A=9F=E8=83=BD=EF=BC=8C=E5=9C=A8=E2=80=9C?= =?UTF-8?q?insert=E2=80=9D=E5=87=BD=E6=95=B0=E4=B8=AD=E5=A6=82=E6=9E=9C?= =?UTF-8?q?=E5=A2=9E=E5=8A=A0=E5=8F=82=E6=95=B0split=5Fby=5Fcharacter?= =?UTF-8?q?=EF=BC=8C=E5=88=99=E4=BC=9A=E6=8C=89=E7=85=A7split=5Fby=5Fchara?= =?UTF-8?q?cter=E8=BF=9B=E8=A1=8C=E5=AD=97=E7=AC=A6=E5=88=86=E5=89=B2?= =?UTF-8?q?=EF=BC=8C=E6=AD=A4=E6=97=B6=E5=A6=82=E6=9E=9C=E6=AF=8F=E4=B8=AA?= =?UTF-8?q?=E5=88=86=E5=89=B2=E5=90=8E=E7=9A=84chunk=E7=9A=84tokens?= =?UTF-8?q?=E5=A4=A7=E4=BA=8Emax=5Ftoken=5Fsize=EF=BC=8C=E5=88=99=E4=BC=9A?= =?UTF-8?q?=E7=BB=A7=E7=BB=AD=E6=8C=89token=5Fsize=E5=88=86=E5=89=B2?= =?UTF-8?q?=EF=BC=88todo=EF=BC=9A=E8=80=83=E8=99=91=E5=AD=97=E7=AC=A6?= =?UTF-8?q?=E5=88=86=E5=89=B2=E5=90=8E=E8=BF=87=E7=9F=AD=E7=9A=84chunk?= =?UTF-8?q?=E5=A4=84=E7=90=86=EF=BC=89?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- lightrag/lightrag.py | 41 ++++--- lightrag/operate.py | 276 +++++++++++++++++++++++-------------------- 2 files changed, 171 insertions(+), 146 deletions(-) diff --git a/lightrag/lightrag.py b/lightrag/lightrag.py index cbe49da2..47d64ac0 100644 --- a/lightrag/lightrag.py +++ b/lightrag/lightrag.py @@ -45,6 +45,7 @@ from .storage import ( from .prompt import GRAPH_FIELD_SEP + # future KG integrations # from .kg.ArangoDB_impl import ( @@ -167,7 +168,7 @@ class LightRAG: # LLM llm_model_func: callable = gpt_4o_mini_complete # hf_model_complete# - llm_model_name: str = "meta-llama/Llama-3.2-1B-Instruct" #'meta-llama/Llama-3.2-1B'#'google/gemma-2-2b-it' + llm_model_name: str = "meta-llama/Llama-3.2-1B-Instruct" # 'meta-llama/Llama-3.2-1B'#'google/gemma-2-2b-it' llm_model_max_token_size: int = 32768 llm_model_max_async: int = 16 llm_model_kwargs: dict = field(default_factory=dict) @@ -267,7 +268,7 @@ class LightRAG: self.llm_model_func, hashing_kv=self.llm_response_cache if self.llm_response_cache - and hasattr(self.llm_response_cache, "global_config") + and hasattr(self.llm_response_cache, "global_config") else self.key_string_value_json_storage_cls( namespace="llm_response_cache", global_config=asdict(self), @@ -313,15 +314,16 @@ class LightRAG: "JsonDocStatusStorage": JsonDocStatusStorage, } - def insert(self, string_or_strings): + def insert(self, string_or_strings, split_by_character=None): loop = always_get_an_event_loop() - return loop.run_until_complete(self.ainsert(string_or_strings)) + return loop.run_until_complete(self.ainsert(string_or_strings, split_by_character)) - async def ainsert(self, string_or_strings): + async def ainsert(self, string_or_strings, split_by_character): """Insert documents with checkpoint support Args: string_or_strings: Single document string or list of document strings + split_by_character: if split_by_character is not None, split the string by character """ if isinstance(string_or_strings, str): string_or_strings = [string_or_strings] @@ -355,10 +357,10 @@ class LightRAG: # Process documents in batches batch_size = self.addon_params.get("insert_batch_size", 10) for i in range(0, len(new_docs), batch_size): - batch_docs = dict(list(new_docs.items())[i : i + batch_size]) + batch_docs = dict(list(new_docs.items())[i: i + batch_size]) for doc_id, doc in tqdm_async( - batch_docs.items(), desc=f"Processing batch {i//batch_size + 1}" + batch_docs.items(), desc=f"Processing batch {i // batch_size + 1}" ): try: # Update status to processing @@ -379,6 +381,7 @@ class LightRAG: } for dp in chunking_by_token_size( doc["content"], + split_by_character=split_by_character, overlap_token_size=self.chunk_overlap_token_size, max_token_size=self.chunk_token_size, tiktoken_model=self.tiktoken_model_name, @@ -545,7 +548,7 @@ class LightRAG: # Check if nodes exist in the knowledge graph for need_insert_id in [src_id, tgt_id]: if not ( - await self.chunk_entity_relation_graph.has_node(need_insert_id) + await self.chunk_entity_relation_graph.has_node(need_insert_id) ): await self.chunk_entity_relation_graph.upsert_node( need_insert_id, @@ -594,9 +597,9 @@ class LightRAG: "src_id": dp["src_id"], "tgt_id": dp["tgt_id"], "content": dp["keywords"] - + dp["src_id"] - + dp["tgt_id"] - + dp["description"], + + dp["src_id"] + + dp["tgt_id"] + + dp["description"], } for dp in all_relationships_data } @@ -621,7 +624,7 @@ class LightRAG: asdict(self), hashing_kv=self.llm_response_cache if self.llm_response_cache - and hasattr(self.llm_response_cache, "global_config") + and hasattr(self.llm_response_cache, "global_config") else self.key_string_value_json_storage_cls( namespace="llm_response_cache", global_config=asdict(self), @@ -637,7 +640,7 @@ class LightRAG: asdict(self), hashing_kv=self.llm_response_cache if self.llm_response_cache - and hasattr(self.llm_response_cache, "global_config") + and hasattr(self.llm_response_cache, "global_config") else self.key_string_value_json_storage_cls( namespace="llm_response_cache", global_config=asdict(self), @@ -656,7 +659,7 @@ class LightRAG: asdict(self), hashing_kv=self.llm_response_cache if self.llm_response_cache - and hasattr(self.llm_response_cache, "global_config") + and hasattr(self.llm_response_cache, "global_config") else self.key_string_value_json_storage_cls( namespace="llm_response_cache", global_config=asdict(self), @@ -897,7 +900,7 @@ class LightRAG: dp for dp in self.entities_vdb.client_storage["data"] if chunk_id - in (dp.get("source_id") or "").split(GRAPH_FIELD_SEP) + in (dp.get("source_id") or "").split(GRAPH_FIELD_SEP) ] if entities_with_chunk: logger.error( @@ -909,7 +912,7 @@ class LightRAG: dp for dp in self.relationships_vdb.client_storage["data"] if chunk_id - in (dp.get("source_id") or "").split(GRAPH_FIELD_SEP) + in (dp.get("source_id") or "").split(GRAPH_FIELD_SEP) ] if relations_with_chunk: logger.error( @@ -926,7 +929,7 @@ class LightRAG: return asyncio.run(self.adelete_by_doc_id(doc_id)) async def get_entity_info( - self, entity_name: str, include_vector_data: bool = False + self, entity_name: str, include_vector_data: bool = False ): """Get detailed information of an entity @@ -977,7 +980,7 @@ class LightRAG: tracemalloc.stop() async def get_relation_info( - self, src_entity: str, tgt_entity: str, include_vector_data: bool = False + self, src_entity: str, tgt_entity: str, include_vector_data: bool = False ): """Get detailed information of a relationship @@ -1019,7 +1022,7 @@ class LightRAG: return result def get_relation_info_sync( - self, src_entity: str, tgt_entity: str, include_vector_data: bool = False + self, src_entity: str, tgt_entity: str, include_vector_data: bool = False ): """Synchronous version of getting relationship information diff --git a/lightrag/operate.py b/lightrag/operate.py index b2c4d215..e8f0df65 100644 --- a/lightrag/operate.py +++ b/lightrag/operate.py @@ -34,30 +34,52 @@ import time def chunking_by_token_size( - content: str, overlap_token_size=128, max_token_size=1024, tiktoken_model="gpt-4o" + content: str, split_by_character=None, overlap_token_size=128, max_token_size=1024, tiktoken_model="gpt-4o" ): tokens = encode_string_by_tiktoken(content, model_name=tiktoken_model) results = [] - for index, start in enumerate( - range(0, len(tokens), max_token_size - overlap_token_size) - ): - chunk_content = decode_tokens_by_tiktoken( - tokens[start : start + max_token_size], model_name=tiktoken_model - ) - results.append( - { - "tokens": min(max_token_size, len(tokens) - start), - "content": chunk_content.strip(), - "chunk_order_index": index, - } - ) + if split_by_character: + raw_chunks = content.split(split_by_character) + new_chunks = [] + for chunk in raw_chunks: + _tokens = encode_string_by_tiktoken(chunk, model_name=tiktoken_model) + if len(_tokens) > max_token_size: + for start in range(0, len(_tokens), max_token_size - overlap_token_size): + chunk_content = decode_tokens_by_tiktoken( + _tokens[start: start + max_token_size], model_name=tiktoken_model + ) + new_chunks.append((min(max_token_size, len(_tokens) - start), chunk_content)) + else: + new_chunks.append((len(_tokens), chunk)) + for index, (_len, chunk) in enumerate(new_chunks): + results.append( + { + "tokens": _len, + "content": chunk.strip(), + "chunk_order_index": index, + } + ) + else: + for index, start in enumerate( + range(0, len(tokens), max_token_size - overlap_token_size) + ): + chunk_content = decode_tokens_by_tiktoken( + tokens[start: start + max_token_size], model_name=tiktoken_model + ) + results.append( + { + "tokens": min(max_token_size, len(tokens) - start), + "content": chunk_content.strip(), + "chunk_order_index": index, + } + ) return results async def _handle_entity_relation_summary( - entity_or_relation_name: str, - description: str, - global_config: dict, + entity_or_relation_name: str, + description: str, + global_config: dict, ) -> str: use_llm_func: callable = global_config["llm_model_func"] llm_max_tokens = global_config["llm_model_max_token_size"] @@ -86,8 +108,8 @@ async def _handle_entity_relation_summary( async def _handle_single_entity_extraction( - record_attributes: list[str], - chunk_key: str, + record_attributes: list[str], + chunk_key: str, ): if len(record_attributes) < 4 or record_attributes[0] != '"entity"': return None @@ -107,8 +129,8 @@ async def _handle_single_entity_extraction( async def _handle_single_relationship_extraction( - record_attributes: list[str], - chunk_key: str, + record_attributes: list[str], + chunk_key: str, ): if len(record_attributes) < 5 or record_attributes[0] != '"relationship"': return None @@ -134,10 +156,10 @@ async def _handle_single_relationship_extraction( async def _merge_nodes_then_upsert( - entity_name: str, - nodes_data: list[dict], - knowledge_graph_inst: BaseGraphStorage, - global_config: dict, + entity_name: str, + nodes_data: list[dict], + knowledge_graph_inst: BaseGraphStorage, + global_config: dict, ): already_entity_types = [] already_source_ids = [] @@ -181,11 +203,11 @@ async def _merge_nodes_then_upsert( async def _merge_edges_then_upsert( - src_id: str, - tgt_id: str, - edges_data: list[dict], - knowledge_graph_inst: BaseGraphStorage, - global_config: dict, + src_id: str, + tgt_id: str, + edges_data: list[dict], + knowledge_graph_inst: BaseGraphStorage, + global_config: dict, ): already_weights = [] already_source_ids = [] @@ -248,12 +270,12 @@ async def _merge_edges_then_upsert( async def extract_entities( - chunks: dict[str, TextChunkSchema], - knowledge_graph_inst: BaseGraphStorage, - entity_vdb: BaseVectorStorage, - relationships_vdb: BaseVectorStorage, - global_config: dict, - llm_response_cache: BaseKVStorage = None, + chunks: dict[str, TextChunkSchema], + knowledge_graph_inst: BaseGraphStorage, + entity_vdb: BaseVectorStorage, + relationships_vdb: BaseVectorStorage, + global_config: dict, + llm_response_cache: BaseKVStorage = None, ) -> Union[BaseGraphStorage, None]: use_llm_func: callable = global_config["llm_model_func"] entity_extract_max_gleaning = global_config["entity_extract_max_gleaning"] @@ -305,13 +327,13 @@ async def extract_entities( already_relations = 0 async def _user_llm_func_with_cache( - input_text: str, history_messages: list[dict[str, str]] = None + input_text: str, history_messages: list[dict[str, str]] = None ) -> str: if enable_llm_cache_for_entity_extract and llm_response_cache: need_to_restore = False if ( - global_config["embedding_cache_config"] - and global_config["embedding_cache_config"]["enabled"] + global_config["embedding_cache_config"] + and global_config["embedding_cache_config"]["enabled"] ): new_config = global_config.copy() new_config["embedding_cache_config"] = None @@ -413,7 +435,7 @@ async def extract_entities( already_relations += len(maybe_edges) now_ticks = PROMPTS["process_tickers"][ already_processed % len(PROMPTS["process_tickers"]) - ] + ] print( f"{now_ticks} Processed {already_processed} chunks, {already_entities} entities(duplicated), {already_relations} relations(duplicated)\r", end="", @@ -423,10 +445,10 @@ async def extract_entities( results = [] for result in tqdm_async( - asyncio.as_completed([_process_single_content(c) for c in ordered_chunks]), - total=len(ordered_chunks), - desc="Extracting entities from chunks", - unit="chunk", + asyncio.as_completed([_process_single_content(c) for c in ordered_chunks]), + total=len(ordered_chunks), + desc="Extracting entities from chunks", + unit="chunk", ): results.append(await result) @@ -440,32 +462,32 @@ async def extract_entities( logger.info("Inserting entities into storage...") all_entities_data = [] for result in tqdm_async( - asyncio.as_completed( - [ - _merge_nodes_then_upsert(k, v, knowledge_graph_inst, global_config) - for k, v in maybe_nodes.items() - ] - ), - total=len(maybe_nodes), - desc="Inserting entities", - unit="entity", + asyncio.as_completed( + [ + _merge_nodes_then_upsert(k, v, knowledge_graph_inst, global_config) + for k, v in maybe_nodes.items() + ] + ), + total=len(maybe_nodes), + desc="Inserting entities", + unit="entity", ): all_entities_data.append(await result) logger.info("Inserting relationships into storage...") all_relationships_data = [] for result in tqdm_async( - asyncio.as_completed( - [ - _merge_edges_then_upsert( - k[0], k[1], v, knowledge_graph_inst, global_config - ) - for k, v in maybe_edges.items() - ] - ), - total=len(maybe_edges), - desc="Inserting relationships", - unit="relationship", + asyncio.as_completed( + [ + _merge_edges_then_upsert( + k[0], k[1], v, knowledge_graph_inst, global_config + ) + for k, v in maybe_edges.items() + ] + ), + total=len(maybe_edges), + desc="Inserting relationships", + unit="relationship", ): all_relationships_data.append(await result) @@ -496,9 +518,9 @@ async def extract_entities( "src_id": dp["src_id"], "tgt_id": dp["tgt_id"], "content": dp["keywords"] - + dp["src_id"] - + dp["tgt_id"] - + dp["description"], + + dp["src_id"] + + dp["tgt_id"] + + dp["description"], "metadata": { "created_at": dp.get("metadata", {}).get("created_at", time.time()) }, @@ -511,14 +533,14 @@ async def extract_entities( async def kg_query( - query, - knowledge_graph_inst: BaseGraphStorage, - entities_vdb: BaseVectorStorage, - relationships_vdb: BaseVectorStorage, - text_chunks_db: BaseKVStorage[TextChunkSchema], - query_param: QueryParam, - global_config: dict, - hashing_kv: BaseKVStorage = None, + query, + knowledge_graph_inst: BaseGraphStorage, + entities_vdb: BaseVectorStorage, + relationships_vdb: BaseVectorStorage, + text_chunks_db: BaseKVStorage[TextChunkSchema], + query_param: QueryParam, + global_config: dict, + hashing_kv: BaseKVStorage = None, ) -> str: # Handle cache use_model_func = global_config["llm_model_func"] @@ -638,12 +660,12 @@ async def kg_query( async def _build_query_context( - query: list, - knowledge_graph_inst: BaseGraphStorage, - entities_vdb: BaseVectorStorage, - relationships_vdb: BaseVectorStorage, - text_chunks_db: BaseKVStorage[TextChunkSchema], - query_param: QueryParam, + query: list, + knowledge_graph_inst: BaseGraphStorage, + entities_vdb: BaseVectorStorage, + relationships_vdb: BaseVectorStorage, + text_chunks_db: BaseKVStorage[TextChunkSchema], + query_param: QueryParam, ): # ll_entities_context, ll_relations_context, ll_text_units_context = "", "", "" # hl_entities_context, hl_relations_context, hl_text_units_context = "", "", "" @@ -696,9 +718,9 @@ async def _build_query_context( query_param, ) if ( - hl_entities_context == "" - and hl_relations_context == "" - and hl_text_units_context == "" + hl_entities_context == "" + and hl_relations_context == "" + and hl_text_units_context == "" ): logger.warn("No high level context found. Switching to local mode.") query_param.mode = "local" @@ -737,11 +759,11 @@ async def _build_query_context( async def _get_node_data( - query, - knowledge_graph_inst: BaseGraphStorage, - entities_vdb: BaseVectorStorage, - text_chunks_db: BaseKVStorage[TextChunkSchema], - query_param: QueryParam, + query, + knowledge_graph_inst: BaseGraphStorage, + entities_vdb: BaseVectorStorage, + text_chunks_db: BaseKVStorage[TextChunkSchema], + query_param: QueryParam, ): # get similar entities results = await entities_vdb.query(query, top_k=query_param.top_k) @@ -828,10 +850,10 @@ async def _get_node_data( async def _find_most_related_text_unit_from_entities( - node_datas: list[dict], - query_param: QueryParam, - text_chunks_db: BaseKVStorage[TextChunkSchema], - knowledge_graph_inst: BaseGraphStorage, + node_datas: list[dict], + query_param: QueryParam, + text_chunks_db: BaseKVStorage[TextChunkSchema], + knowledge_graph_inst: BaseGraphStorage, ): text_units = [ split_string_by_multi_markers(dp["source_id"], [GRAPH_FIELD_SEP]) @@ -871,8 +893,8 @@ async def _find_most_related_text_unit_from_entities( if this_edges: for e in this_edges: if ( - e[1] in all_one_hop_text_units_lookup - and c_id in all_one_hop_text_units_lookup[e[1]] + e[1] in all_one_hop_text_units_lookup + and c_id in all_one_hop_text_units_lookup[e[1]] ): all_text_units_lookup[c_id]["relation_counts"] += 1 @@ -902,9 +924,9 @@ async def _find_most_related_text_unit_from_entities( async def _find_most_related_edges_from_entities( - node_datas: list[dict], - query_param: QueryParam, - knowledge_graph_inst: BaseGraphStorage, + node_datas: list[dict], + query_param: QueryParam, + knowledge_graph_inst: BaseGraphStorage, ): all_related_edges = await asyncio.gather( *[knowledge_graph_inst.get_node_edges(dp["entity_name"]) for dp in node_datas] @@ -942,11 +964,11 @@ async def _find_most_related_edges_from_entities( async def _get_edge_data( - keywords, - knowledge_graph_inst: BaseGraphStorage, - relationships_vdb: BaseVectorStorage, - text_chunks_db: BaseKVStorage[TextChunkSchema], - query_param: QueryParam, + keywords, + knowledge_graph_inst: BaseGraphStorage, + relationships_vdb: BaseVectorStorage, + text_chunks_db: BaseKVStorage[TextChunkSchema], + query_param: QueryParam, ): results = await relationships_vdb.query(keywords, top_k=query_param.top_k) @@ -1044,9 +1066,9 @@ async def _get_edge_data( async def _find_most_related_entities_from_relationships( - edge_datas: list[dict], - query_param: QueryParam, - knowledge_graph_inst: BaseGraphStorage, + edge_datas: list[dict], + query_param: QueryParam, + knowledge_graph_inst: BaseGraphStorage, ): entity_names = [] seen = set() @@ -1081,10 +1103,10 @@ async def _find_most_related_entities_from_relationships( async def _find_related_text_unit_from_relationships( - edge_datas: list[dict], - query_param: QueryParam, - text_chunks_db: BaseKVStorage[TextChunkSchema], - knowledge_graph_inst: BaseGraphStorage, + edge_datas: list[dict], + query_param: QueryParam, + text_chunks_db: BaseKVStorage[TextChunkSchema], + knowledge_graph_inst: BaseGraphStorage, ): text_units = [ split_string_by_multi_markers(dp["source_id"], [GRAPH_FIELD_SEP]) @@ -1150,12 +1172,12 @@ def combine_contexts(entities, relationships, sources): async def naive_query( - query, - chunks_vdb: BaseVectorStorage, - text_chunks_db: BaseKVStorage[TextChunkSchema], - query_param: QueryParam, - global_config: dict, - hashing_kv: BaseKVStorage = None, + query, + chunks_vdb: BaseVectorStorage, + text_chunks_db: BaseKVStorage[TextChunkSchema], + query_param: QueryParam, + global_config: dict, + hashing_kv: BaseKVStorage = None, ): # Handle cache use_model_func = global_config["llm_model_func"] @@ -1213,7 +1235,7 @@ async def naive_query( if len(response) > len(sys_prompt): response = ( - response[len(sys_prompt) :] + response[len(sys_prompt):] .replace(sys_prompt, "") .replace("user", "") .replace("model", "") @@ -1241,15 +1263,15 @@ async def naive_query( async def mix_kg_vector_query( - query, - knowledge_graph_inst: BaseGraphStorage, - entities_vdb: BaseVectorStorage, - relationships_vdb: BaseVectorStorage, - chunks_vdb: BaseVectorStorage, - text_chunks_db: BaseKVStorage[TextChunkSchema], - query_param: QueryParam, - global_config: dict, - hashing_kv: BaseKVStorage = None, + query, + knowledge_graph_inst: BaseGraphStorage, + entities_vdb: BaseVectorStorage, + relationships_vdb: BaseVectorStorage, + chunks_vdb: BaseVectorStorage, + text_chunks_db: BaseKVStorage[TextChunkSchema], + query_param: QueryParam, + global_config: dict, + hashing_kv: BaseKVStorage = None, ) -> str: """ Hybrid retrieval implementation combining knowledge graph and vector search. @@ -1274,7 +1296,7 @@ async def mix_kg_vector_query( # Reuse keyword extraction logic from kg_query example_number = global_config["addon_params"].get("example_number", None) if example_number and example_number < len( - PROMPTS["keywords_extraction_examples"] + PROMPTS["keywords_extraction_examples"] ): examples = "\n".join( PROMPTS["keywords_extraction_examples"][: int(example_number)] From 3bbd3ee1b232cf1335617a5f4308651b295061b5 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?=E7=AB=A5=E7=9F=B3=E6=B8=8A?= Date: Tue, 7 Jan 2025 13:45:18 +0800 Subject: [PATCH 2/3] =?UTF-8?q?=E5=9C=A8Mac=E7=AB=AFtorch~=3D2.5.1+cu121?= =?UTF-8?q?=E4=BC=9A=E5=AF=BC=E8=87=B4=E6=9C=AC=E5=9C=B0=E5=AE=89=E8=A3=85?= =?UTF-8?q?=E6=97=B6=E6=8A=A5=E9=94=99?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- requirements.txt | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/requirements.txt b/requirements.txt index 79249e7e..dd3c4cf3 100644 --- a/requirements.txt +++ b/requirements.txt @@ -32,7 +32,8 @@ tenacity~=9.0.0 # LLM packages tiktoken~=0.8.0 -torch~=2.5.1+cu121 +# torch~=2.5.1+cu121 +torch~=2.5.1 tqdm~=4.67.1 transformers~=4.47.1 xxhash From 6b19401dc6f0a27597f15990bd86206409feb540 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?=E7=AB=A5=E7=9F=B3=E6=B8=8A?= Date: Tue, 7 Jan 2025 16:26:12 +0800 Subject: [PATCH 3/3] chunk split retry --- lightrag/lightrag.py | 34 +- lightrag/operate.py | 247 ++++++++------- test.ipynb | 740 +++++++++++++++++++++++++++++++++++++++++++ 3 files changed, 886 insertions(+), 135 deletions(-) create mode 100644 test.ipynb diff --git a/lightrag/lightrag.py b/lightrag/lightrag.py index 47d64ac0..7496d736 100644 --- a/lightrag/lightrag.py +++ b/lightrag/lightrag.py @@ -268,7 +268,7 @@ class LightRAG: self.llm_model_func, hashing_kv=self.llm_response_cache if self.llm_response_cache - and hasattr(self.llm_response_cache, "global_config") + and hasattr(self.llm_response_cache, "global_config") else self.key_string_value_json_storage_cls( namespace="llm_response_cache", global_config=asdict(self), @@ -316,7 +316,9 @@ class LightRAG: def insert(self, string_or_strings, split_by_character=None): loop = always_get_an_event_loop() - return loop.run_until_complete(self.ainsert(string_or_strings, split_by_character)) + return loop.run_until_complete( + self.ainsert(string_or_strings, split_by_character) + ) async def ainsert(self, string_or_strings, split_by_character): """Insert documents with checkpoint support @@ -357,10 +359,10 @@ class LightRAG: # Process documents in batches batch_size = self.addon_params.get("insert_batch_size", 10) for i in range(0, len(new_docs), batch_size): - batch_docs = dict(list(new_docs.items())[i: i + batch_size]) + batch_docs = dict(list(new_docs.items())[i : i + batch_size]) for doc_id, doc in tqdm_async( - batch_docs.items(), desc=f"Processing batch {i // batch_size + 1}" + batch_docs.items(), desc=f"Processing batch {i // batch_size + 1}" ): try: # Update status to processing @@ -548,7 +550,7 @@ class LightRAG: # Check if nodes exist in the knowledge graph for need_insert_id in [src_id, tgt_id]: if not ( - await self.chunk_entity_relation_graph.has_node(need_insert_id) + await self.chunk_entity_relation_graph.has_node(need_insert_id) ): await self.chunk_entity_relation_graph.upsert_node( need_insert_id, @@ -597,9 +599,9 @@ class LightRAG: "src_id": dp["src_id"], "tgt_id": dp["tgt_id"], "content": dp["keywords"] - + dp["src_id"] - + dp["tgt_id"] - + dp["description"], + + dp["src_id"] + + dp["tgt_id"] + + dp["description"], } for dp in all_relationships_data } @@ -624,7 +626,7 @@ class LightRAG: asdict(self), hashing_kv=self.llm_response_cache if self.llm_response_cache - and hasattr(self.llm_response_cache, "global_config") + and hasattr(self.llm_response_cache, "global_config") else self.key_string_value_json_storage_cls( namespace="llm_response_cache", global_config=asdict(self), @@ -640,7 +642,7 @@ class LightRAG: asdict(self), hashing_kv=self.llm_response_cache if self.llm_response_cache - and hasattr(self.llm_response_cache, "global_config") + and hasattr(self.llm_response_cache, "global_config") else self.key_string_value_json_storage_cls( namespace="llm_response_cache", global_config=asdict(self), @@ -659,7 +661,7 @@ class LightRAG: asdict(self), hashing_kv=self.llm_response_cache if self.llm_response_cache - and hasattr(self.llm_response_cache, "global_config") + and hasattr(self.llm_response_cache, "global_config") else self.key_string_value_json_storage_cls( namespace="llm_response_cache", global_config=asdict(self), @@ -900,7 +902,7 @@ class LightRAG: dp for dp in self.entities_vdb.client_storage["data"] if chunk_id - in (dp.get("source_id") or "").split(GRAPH_FIELD_SEP) + in (dp.get("source_id") or "").split(GRAPH_FIELD_SEP) ] if entities_with_chunk: logger.error( @@ -912,7 +914,7 @@ class LightRAG: dp for dp in self.relationships_vdb.client_storage["data"] if chunk_id - in (dp.get("source_id") or "").split(GRAPH_FIELD_SEP) + in (dp.get("source_id") or "").split(GRAPH_FIELD_SEP) ] if relations_with_chunk: logger.error( @@ -929,7 +931,7 @@ class LightRAG: return asyncio.run(self.adelete_by_doc_id(doc_id)) async def get_entity_info( - self, entity_name: str, include_vector_data: bool = False + self, entity_name: str, include_vector_data: bool = False ): """Get detailed information of an entity @@ -980,7 +982,7 @@ class LightRAG: tracemalloc.stop() async def get_relation_info( - self, src_entity: str, tgt_entity: str, include_vector_data: bool = False + self, src_entity: str, tgt_entity: str, include_vector_data: bool = False ): """Get detailed information of a relationship @@ -1022,7 +1024,7 @@ class LightRAG: return result def get_relation_info_sync( - self, src_entity: str, tgt_entity: str, include_vector_data: bool = False + self, src_entity: str, tgt_entity: str, include_vector_data: bool = False ): """Synchronous version of getting relationship information diff --git a/lightrag/operate.py b/lightrag/operate.py index e8f0df65..1128b41c 100644 --- a/lightrag/operate.py +++ b/lightrag/operate.py @@ -34,7 +34,11 @@ import time def chunking_by_token_size( - content: str, split_by_character=None, overlap_token_size=128, max_token_size=1024, tiktoken_model="gpt-4o" + content: str, + split_by_character=None, + overlap_token_size=128, + max_token_size=1024, + tiktoken_model="gpt-4o", ): tokens = encode_string_by_tiktoken(content, model_name=tiktoken_model) results = [] @@ -44,11 +48,16 @@ def chunking_by_token_size( for chunk in raw_chunks: _tokens = encode_string_by_tiktoken(chunk, model_name=tiktoken_model) if len(_tokens) > max_token_size: - for start in range(0, len(_tokens), max_token_size - overlap_token_size): + for start in range( + 0, len(_tokens), max_token_size - overlap_token_size + ): chunk_content = decode_tokens_by_tiktoken( - _tokens[start: start + max_token_size], model_name=tiktoken_model + _tokens[start : start + max_token_size], + model_name=tiktoken_model, + ) + new_chunks.append( + (min(max_token_size, len(_tokens) - start), chunk_content) ) - new_chunks.append((min(max_token_size, len(_tokens) - start), chunk_content)) else: new_chunks.append((len(_tokens), chunk)) for index, (_len, chunk) in enumerate(new_chunks): @@ -61,10 +70,10 @@ def chunking_by_token_size( ) else: for index, start in enumerate( - range(0, len(tokens), max_token_size - overlap_token_size) + range(0, len(tokens), max_token_size - overlap_token_size) ): chunk_content = decode_tokens_by_tiktoken( - tokens[start: start + max_token_size], model_name=tiktoken_model + tokens[start : start + max_token_size], model_name=tiktoken_model ) results.append( { @@ -77,9 +86,9 @@ def chunking_by_token_size( async def _handle_entity_relation_summary( - entity_or_relation_name: str, - description: str, - global_config: dict, + entity_or_relation_name: str, + description: str, + global_config: dict, ) -> str: use_llm_func: callable = global_config["llm_model_func"] llm_max_tokens = global_config["llm_model_max_token_size"] @@ -108,8 +117,8 @@ async def _handle_entity_relation_summary( async def _handle_single_entity_extraction( - record_attributes: list[str], - chunk_key: str, + record_attributes: list[str], + chunk_key: str, ): if len(record_attributes) < 4 or record_attributes[0] != '"entity"': return None @@ -129,8 +138,8 @@ async def _handle_single_entity_extraction( async def _handle_single_relationship_extraction( - record_attributes: list[str], - chunk_key: str, + record_attributes: list[str], + chunk_key: str, ): if len(record_attributes) < 5 or record_attributes[0] != '"relationship"': return None @@ -156,10 +165,10 @@ async def _handle_single_relationship_extraction( async def _merge_nodes_then_upsert( - entity_name: str, - nodes_data: list[dict], - knowledge_graph_inst: BaseGraphStorage, - global_config: dict, + entity_name: str, + nodes_data: list[dict], + knowledge_graph_inst: BaseGraphStorage, + global_config: dict, ): already_entity_types = [] already_source_ids = [] @@ -203,11 +212,11 @@ async def _merge_nodes_then_upsert( async def _merge_edges_then_upsert( - src_id: str, - tgt_id: str, - edges_data: list[dict], - knowledge_graph_inst: BaseGraphStorage, - global_config: dict, + src_id: str, + tgt_id: str, + edges_data: list[dict], + knowledge_graph_inst: BaseGraphStorage, + global_config: dict, ): already_weights = [] already_source_ids = [] @@ -270,12 +279,12 @@ async def _merge_edges_then_upsert( async def extract_entities( - chunks: dict[str, TextChunkSchema], - knowledge_graph_inst: BaseGraphStorage, - entity_vdb: BaseVectorStorage, - relationships_vdb: BaseVectorStorage, - global_config: dict, - llm_response_cache: BaseKVStorage = None, + chunks: dict[str, TextChunkSchema], + knowledge_graph_inst: BaseGraphStorage, + entity_vdb: BaseVectorStorage, + relationships_vdb: BaseVectorStorage, + global_config: dict, + llm_response_cache: BaseKVStorage = None, ) -> Union[BaseGraphStorage, None]: use_llm_func: callable = global_config["llm_model_func"] entity_extract_max_gleaning = global_config["entity_extract_max_gleaning"] @@ -327,13 +336,13 @@ async def extract_entities( already_relations = 0 async def _user_llm_func_with_cache( - input_text: str, history_messages: list[dict[str, str]] = None + input_text: str, history_messages: list[dict[str, str]] = None ) -> str: if enable_llm_cache_for_entity_extract and llm_response_cache: need_to_restore = False if ( - global_config["embedding_cache_config"] - and global_config["embedding_cache_config"]["enabled"] + global_config["embedding_cache_config"] + and global_config["embedding_cache_config"]["enabled"] ): new_config = global_config.copy() new_config["embedding_cache_config"] = None @@ -435,7 +444,7 @@ async def extract_entities( already_relations += len(maybe_edges) now_ticks = PROMPTS["process_tickers"][ already_processed % len(PROMPTS["process_tickers"]) - ] + ] print( f"{now_ticks} Processed {already_processed} chunks, {already_entities} entities(duplicated), {already_relations} relations(duplicated)\r", end="", @@ -445,10 +454,10 @@ async def extract_entities( results = [] for result in tqdm_async( - asyncio.as_completed([_process_single_content(c) for c in ordered_chunks]), - total=len(ordered_chunks), - desc="Extracting entities from chunks", - unit="chunk", + asyncio.as_completed([_process_single_content(c) for c in ordered_chunks]), + total=len(ordered_chunks), + desc="Extracting entities from chunks", + unit="chunk", ): results.append(await result) @@ -462,32 +471,32 @@ async def extract_entities( logger.info("Inserting entities into storage...") all_entities_data = [] for result in tqdm_async( - asyncio.as_completed( - [ - _merge_nodes_then_upsert(k, v, knowledge_graph_inst, global_config) - for k, v in maybe_nodes.items() - ] - ), - total=len(maybe_nodes), - desc="Inserting entities", - unit="entity", + asyncio.as_completed( + [ + _merge_nodes_then_upsert(k, v, knowledge_graph_inst, global_config) + for k, v in maybe_nodes.items() + ] + ), + total=len(maybe_nodes), + desc="Inserting entities", + unit="entity", ): all_entities_data.append(await result) logger.info("Inserting relationships into storage...") all_relationships_data = [] for result in tqdm_async( - asyncio.as_completed( - [ - _merge_edges_then_upsert( - k[0], k[1], v, knowledge_graph_inst, global_config - ) - for k, v in maybe_edges.items() - ] - ), - total=len(maybe_edges), - desc="Inserting relationships", - unit="relationship", + asyncio.as_completed( + [ + _merge_edges_then_upsert( + k[0], k[1], v, knowledge_graph_inst, global_config + ) + for k, v in maybe_edges.items() + ] + ), + total=len(maybe_edges), + desc="Inserting relationships", + unit="relationship", ): all_relationships_data.append(await result) @@ -518,9 +527,9 @@ async def extract_entities( "src_id": dp["src_id"], "tgt_id": dp["tgt_id"], "content": dp["keywords"] - + dp["src_id"] - + dp["tgt_id"] - + dp["description"], + + dp["src_id"] + + dp["tgt_id"] + + dp["description"], "metadata": { "created_at": dp.get("metadata", {}).get("created_at", time.time()) }, @@ -533,14 +542,14 @@ async def extract_entities( async def kg_query( - query, - knowledge_graph_inst: BaseGraphStorage, - entities_vdb: BaseVectorStorage, - relationships_vdb: BaseVectorStorage, - text_chunks_db: BaseKVStorage[TextChunkSchema], - query_param: QueryParam, - global_config: dict, - hashing_kv: BaseKVStorage = None, + query, + knowledge_graph_inst: BaseGraphStorage, + entities_vdb: BaseVectorStorage, + relationships_vdb: BaseVectorStorage, + text_chunks_db: BaseKVStorage[TextChunkSchema], + query_param: QueryParam, + global_config: dict, + hashing_kv: BaseKVStorage = None, ) -> str: # Handle cache use_model_func = global_config["llm_model_func"] @@ -660,12 +669,12 @@ async def kg_query( async def _build_query_context( - query: list, - knowledge_graph_inst: BaseGraphStorage, - entities_vdb: BaseVectorStorage, - relationships_vdb: BaseVectorStorage, - text_chunks_db: BaseKVStorage[TextChunkSchema], - query_param: QueryParam, + query: list, + knowledge_graph_inst: BaseGraphStorage, + entities_vdb: BaseVectorStorage, + relationships_vdb: BaseVectorStorage, + text_chunks_db: BaseKVStorage[TextChunkSchema], + query_param: QueryParam, ): # ll_entities_context, ll_relations_context, ll_text_units_context = "", "", "" # hl_entities_context, hl_relations_context, hl_text_units_context = "", "", "" @@ -718,9 +727,9 @@ async def _build_query_context( query_param, ) if ( - hl_entities_context == "" - and hl_relations_context == "" - and hl_text_units_context == "" + hl_entities_context == "" + and hl_relations_context == "" + and hl_text_units_context == "" ): logger.warn("No high level context found. Switching to local mode.") query_param.mode = "local" @@ -759,11 +768,11 @@ async def _build_query_context( async def _get_node_data( - query, - knowledge_graph_inst: BaseGraphStorage, - entities_vdb: BaseVectorStorage, - text_chunks_db: BaseKVStorage[TextChunkSchema], - query_param: QueryParam, + query, + knowledge_graph_inst: BaseGraphStorage, + entities_vdb: BaseVectorStorage, + text_chunks_db: BaseKVStorage[TextChunkSchema], + query_param: QueryParam, ): # get similar entities results = await entities_vdb.query(query, top_k=query_param.top_k) @@ -850,10 +859,10 @@ async def _get_node_data( async def _find_most_related_text_unit_from_entities( - node_datas: list[dict], - query_param: QueryParam, - text_chunks_db: BaseKVStorage[TextChunkSchema], - knowledge_graph_inst: BaseGraphStorage, + node_datas: list[dict], + query_param: QueryParam, + text_chunks_db: BaseKVStorage[TextChunkSchema], + knowledge_graph_inst: BaseGraphStorage, ): text_units = [ split_string_by_multi_markers(dp["source_id"], [GRAPH_FIELD_SEP]) @@ -893,8 +902,8 @@ async def _find_most_related_text_unit_from_entities( if this_edges: for e in this_edges: if ( - e[1] in all_one_hop_text_units_lookup - and c_id in all_one_hop_text_units_lookup[e[1]] + e[1] in all_one_hop_text_units_lookup + and c_id in all_one_hop_text_units_lookup[e[1]] ): all_text_units_lookup[c_id]["relation_counts"] += 1 @@ -924,9 +933,9 @@ async def _find_most_related_text_unit_from_entities( async def _find_most_related_edges_from_entities( - node_datas: list[dict], - query_param: QueryParam, - knowledge_graph_inst: BaseGraphStorage, + node_datas: list[dict], + query_param: QueryParam, + knowledge_graph_inst: BaseGraphStorage, ): all_related_edges = await asyncio.gather( *[knowledge_graph_inst.get_node_edges(dp["entity_name"]) for dp in node_datas] @@ -964,11 +973,11 @@ async def _find_most_related_edges_from_entities( async def _get_edge_data( - keywords, - knowledge_graph_inst: BaseGraphStorage, - relationships_vdb: BaseVectorStorage, - text_chunks_db: BaseKVStorage[TextChunkSchema], - query_param: QueryParam, + keywords, + knowledge_graph_inst: BaseGraphStorage, + relationships_vdb: BaseVectorStorage, + text_chunks_db: BaseKVStorage[TextChunkSchema], + query_param: QueryParam, ): results = await relationships_vdb.query(keywords, top_k=query_param.top_k) @@ -1066,9 +1075,9 @@ async def _get_edge_data( async def _find_most_related_entities_from_relationships( - edge_datas: list[dict], - query_param: QueryParam, - knowledge_graph_inst: BaseGraphStorage, + edge_datas: list[dict], + query_param: QueryParam, + knowledge_graph_inst: BaseGraphStorage, ): entity_names = [] seen = set() @@ -1103,10 +1112,10 @@ async def _find_most_related_entities_from_relationships( async def _find_related_text_unit_from_relationships( - edge_datas: list[dict], - query_param: QueryParam, - text_chunks_db: BaseKVStorage[TextChunkSchema], - knowledge_graph_inst: BaseGraphStorage, + edge_datas: list[dict], + query_param: QueryParam, + text_chunks_db: BaseKVStorage[TextChunkSchema], + knowledge_graph_inst: BaseGraphStorage, ): text_units = [ split_string_by_multi_markers(dp["source_id"], [GRAPH_FIELD_SEP]) @@ -1172,12 +1181,12 @@ def combine_contexts(entities, relationships, sources): async def naive_query( - query, - chunks_vdb: BaseVectorStorage, - text_chunks_db: BaseKVStorage[TextChunkSchema], - query_param: QueryParam, - global_config: dict, - hashing_kv: BaseKVStorage = None, + query, + chunks_vdb: BaseVectorStorage, + text_chunks_db: BaseKVStorage[TextChunkSchema], + query_param: QueryParam, + global_config: dict, + hashing_kv: BaseKVStorage = None, ): # Handle cache use_model_func = global_config["llm_model_func"] @@ -1235,7 +1244,7 @@ async def naive_query( if len(response) > len(sys_prompt): response = ( - response[len(sys_prompt):] + response[len(sys_prompt) :] .replace(sys_prompt, "") .replace("user", "") .replace("model", "") @@ -1263,15 +1272,15 @@ async def naive_query( async def mix_kg_vector_query( - query, - knowledge_graph_inst: BaseGraphStorage, - entities_vdb: BaseVectorStorage, - relationships_vdb: BaseVectorStorage, - chunks_vdb: BaseVectorStorage, - text_chunks_db: BaseKVStorage[TextChunkSchema], - query_param: QueryParam, - global_config: dict, - hashing_kv: BaseKVStorage = None, + query, + knowledge_graph_inst: BaseGraphStorage, + entities_vdb: BaseVectorStorage, + relationships_vdb: BaseVectorStorage, + chunks_vdb: BaseVectorStorage, + text_chunks_db: BaseKVStorage[TextChunkSchema], + query_param: QueryParam, + global_config: dict, + hashing_kv: BaseKVStorage = None, ) -> str: """ Hybrid retrieval implementation combining knowledge graph and vector search. @@ -1296,7 +1305,7 @@ async def mix_kg_vector_query( # Reuse keyword extraction logic from kg_query example_number = global_config["addon_params"].get("example_number", None) if example_number and example_number < len( - PROMPTS["keywords_extraction_examples"] + PROMPTS["keywords_extraction_examples"] ): examples = "\n".join( PROMPTS["keywords_extraction_examples"][: int(example_number)] diff --git a/test.ipynb b/test.ipynb new file mode 100644 index 00000000..2b9253b4 --- /dev/null +++ b/test.ipynb @@ -0,0 +1,740 @@ +{ + "cells": [ + { + "cell_type": "code", + "execution_count": 1, + "id": "4b5690db12e34685", + "metadata": { + "ExecuteTime": { + "end_time": "2025-01-07T05:38:34.174205Z", + "start_time": "2025-01-07T05:38:29.978194Z" + } + }, + "outputs": [], + "source": [ + "import os\n", + "import logging\n", + "import numpy as np\n", + "from lightrag import LightRAG, QueryParam\n", + "from lightrag.llm import openai_complete_if_cache, openai_embedding\n", + "from lightrag.utils import EmbeddingFunc\n", + "import nest_asyncio" + ] + }, + { + "cell_type": "code", + "execution_count": 2, + "id": "8c8ee7c061bf9159", + "metadata": { + "ExecuteTime": { + "end_time": "2025-01-07T05:38:37.440083Z", + "start_time": "2025-01-07T05:38:37.437666Z" + } + }, + "outputs": [], + "source": [ + "nest_asyncio.apply()\n", + "WORKING_DIR = \"../llm_rag/paper_db/R000088_test2\"\n", + "logging.basicConfig(format=\"%(levelname)s:%(message)s\", level=logging.INFO)\n", + "if not os.path.exists(WORKING_DIR):\n", + " os.mkdir(WORKING_DIR)\n", + "os.environ[\"doubao_api\"] = \"6b890250-0cf6-4eb1-aa82-9c9d711398a7\"" + ] + }, + { + "cell_type": "code", + "execution_count": 3, + "id": "a5009d16e0851dca", + "metadata": { + "ExecuteTime": { + "end_time": "2025-01-07T05:38:42.594315Z", + "start_time": "2025-01-07T05:38:42.590800Z" + } + }, + "outputs": [], + "source": [ + "async def llm_model_func(\n", + " prompt, system_prompt=None, history_messages=[], keyword_extraction=False, **kwargs\n", + ") -> str:\n", + " return await openai_complete_if_cache(\n", + " \"ep-20241218114828-2tlww\",\n", + " prompt,\n", + " system_prompt=system_prompt,\n", + " history_messages=history_messages,\n", + " api_key=os.getenv(\"doubao_api\"),\n", + " base_url=\"https://ark.cn-beijing.volces.com/api/v3\",\n", + " **kwargs,\n", + " )\n", + "\n", + "\n", + "async def embedding_func(texts: list[str]) -> np.ndarray:\n", + " return await openai_embedding(\n", + " texts,\n", + " model=\"ep-20241231173413-pgjmk\",\n", + " api_key=os.getenv(\"doubao_api\"),\n", + " base_url=\"https://ark.cn-beijing.volces.com/api/v3\",\n", + " )" + ] + }, + { + "cell_type": "code", + "execution_count": 4, + "id": "397fcad24ce4d0ed", + "metadata": { + "ExecuteTime": { + "end_time": "2025-01-07T05:38:44.016901Z", + "start_time": "2025-01-07T05:38:44.006291Z" + } + }, + "outputs": [ + { + "name": "stderr", + "output_type": "stream", + "text": [ + "INFO:lightrag:Logger initialized for working directory: ../llm_rag/paper_db/R000088_test2\n", + "INFO:lightrag:Load KV llm_response_cache with 0 data\n", + "INFO:lightrag:Load KV full_docs with 0 data\n", + "INFO:lightrag:Load KV text_chunks with 0 data\n", + "INFO:nano-vectordb:Init {'embedding_dim': 4096, 'metric': 'cosine', 'storage_file': '../llm_rag/paper_db/R000088_test2/vdb_entities.json'} 0 data\n", + "INFO:nano-vectordb:Init {'embedding_dim': 4096, 'metric': 'cosine', 'storage_file': '../llm_rag/paper_db/R000088_test2/vdb_relationships.json'} 0 data\n", + "INFO:nano-vectordb:Init {'embedding_dim': 4096, 'metric': 'cosine', 'storage_file': '../llm_rag/paper_db/R000088_test2/vdb_chunks.json'} 0 data\n", + "INFO:lightrag:Loaded document status storage with 0 records\n" + ] + } + ], + "source": [ + "rag = LightRAG(\n", + " working_dir=WORKING_DIR,\n", + " llm_model_func=llm_model_func,\n", + " embedding_func=EmbeddingFunc(\n", + " embedding_dim=4096, max_token_size=8192, func=embedding_func\n", + " ),\n", + ")" + ] + }, + { + "cell_type": "code", + "execution_count": 5, + "id": "1dc3603677f7484d", + "metadata": { + "ExecuteTime": { + "end_time": "2025-01-07T05:38:47.509111Z", + "start_time": "2025-01-07T05:38:47.501997Z" + } + }, + "outputs": [], + "source": [ + "with open(\n", + " \"../llm_rag/example/R000088/auto/R000088_full_txt.md\", \"r\", encoding=\"utf-8\"\n", + ") as f:\n", + " content = f.read()\n", + "\n", + "\n", + "async def embedding_func(texts: list[str]) -> np.ndarray:\n", + " return await openai_embedding(\n", + " texts,\n", + " model=\"ep-20241231173413-pgjmk\",\n", + " api_key=os.getenv(\"doubao_api\"),\n", + " base_url=\"https://ark.cn-beijing.volces.com/api/v3\",\n", + " )\n", + "\n", + "\n", + "async def get_embedding_dim():\n", + " test_text = [\"This is a test sentence.\"]\n", + " embedding = await embedding_func(test_text)\n", + " embedding_dim = embedding.shape[1]\n", + " return embedding_dim" + ] + }, + { + "cell_type": "code", + "execution_count": 6, + "id": "6844202606acfbe5", + "metadata": { + "ExecuteTime": { + "end_time": "2025-01-07T05:38:50.666764Z", + "start_time": "2025-01-07T05:38:50.247712Z" + } + }, + "outputs": [ + { + "name": "stderr", + "output_type": "stream", + "text": [ + "INFO:httpx:HTTP Request: POST https://ark.cn-beijing.volces.com/api/v3/embeddings \"HTTP/1.1 200 OK\"\n" + ] + } + ], + "source": [ + "embedding_dimension = await get_embedding_dim()" + ] + }, + { + "cell_type": "code", + "execution_count": 7, + "id": "d6273839d9681403", + "metadata": { + "ExecuteTime": { + "end_time": "2025-01-07T05:42:33.085507Z", + "start_time": "2025-01-07T05:38:56.789348Z" + } + }, + "outputs": [ + { + "name": "stderr", + "output_type": "stream", + "text": [ + "INFO:lightrag:Processing 1 new unique documents\n", + "Processing batch 1: 0%| | 0/1 [00:00标签中，针对每个问题详细分析你的思考过程。然后在<回答>标签中给出所有问题的最终答案。\"\"\"" + ] + }, + { + "cell_type": "code", + "execution_count": 9, + "id": "7a6491385b050095", + "metadata": { + "ExecuteTime": { + "end_time": "2025-01-07T05:43:24.751628Z", + "start_time": "2025-01-07T05:42:50.865679Z" + } + }, + "outputs": [ + { + "name": "stderr", + "output_type": "stream", + "text": [ + "INFO:httpx:HTTP Request: POST https://ark.cn-beijing.volces.com/api/v3/chat/completions \"HTTP/1.1 200 OK\"\n", + "INFO:lightrag:kw_prompt result:\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "{\n", + " \"high_level_keywords\": [\"英文学术研究论文分析\", \"关键信息提取\", \"深入分析\"],\n", + " \"low_level_keywords\": [\"研究队列\", \"队列名称\", \"队列开展国家\", \"性别分布\", \"年龄分布\", \"队列研究时间线\", \"实际参与研究人数\"]\n", + "}\n" + ] + }, + { + "name": "stderr", + "output_type": "stream", + "text": [ + "INFO:httpx:HTTP Request: POST https://ark.cn-beijing.volces.com/api/v3/embeddings \"HTTP/1.1 200 OK\"\n", + "INFO:lightrag:Local query uses 60 entites, 38 relations, 6 text units\n", + "INFO:httpx:HTTP Request: POST https://ark.cn-beijing.volces.com/api/v3/embeddings \"HTTP/1.1 200 OK\"\n", + "INFO:lightrag:Global query uses 72 entites, 60 relations, 4 text units\n", + "INFO:httpx:HTTP Request: POST https://ark.cn-beijing.volces.com/api/v3/chat/completions \"HTTP/1.1 200 OK\"\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "<分析>\n", + "- **分析对象来自哪些研究队列及是单独分析还是联合分析**：\n", + " 通过查找论文内容，发现文中提到“This is a combined analysis of data from 2 randomized, double-blind, placebo-controlled clinical trials (Norwegian Vitamin [NORVIT] trial15 and Western Norway B Vitamin Intervention Trial [WENBIT]16)”，明确是对两个队列的数据进行联合分析，队列名称分别为“Norwegian Vitamin (NORVIT) trial”和“Western Norway B Vitamin Intervention Trial (WENBIT)”。\n", + "- **队列开展的国家**：\n", + " 文中多次提及研究在挪威进行，如“combined analyses and extended follow-up of 2 vitamin B intervention trials among patients with ischemic heart disease in Norway”，所以确定研究开展的国家是挪威。\n", + "- **队列研究对象的性别分布**：\n", + " 从“Mean (SD) age was 62.3 (11.0) years and 23.5% of participants were women”可知，研究对象包含男性和女性，即全体。\n", + "- **队列收集结束时研究对象年龄分布**：\n", + " 已知“Mean (SD) age was 62.3 (11.0) years”是基线时年龄信息，“Median (interquartile range) duration of extended follow-up through December 31, 2007, was 78 (61 - 90) months”，由于随访的中位时间是78个月（约6.5年），所以可推算队列收集结束时研究对象年龄均值约为62.3 + 6.5 = 68.8岁（标准差仍为11.0年）。\n", + "- **队列研究时间线**：\n", + " 根据“2 randomized, double-blind, placebo-controlled clinical trials (Norwegian Vitamin [NORVIT] trial15 and Western Norway B Vitamin Intervention Trial [WENBIT]16) conducted between 1998 and 2005, and an observational posttrial follow-up through December 31, 2007”可知，队列开始收集信息时间为1998年，结束时间为2007年12月31日。\n", + "- **队列结束时实际参与研究人数**：\n", + " 由“A total of 6837 individuals were included in the combined analyses, of whom 6261 (91.6%) participated in posttrial follow-up”可知，队列结束时实际参与研究人数为6261人。\n", + "\n", + "\n", + "<回答>\n", + "- 分析对象来自“Norwegian Vitamin (NORVIT) trial”和“Western Norway B Vitamin Intervention Trial (WENBIT)”两个研究队列，文中是对这两个队列的数据进行联合分析。\n", + "- 队列开展的国家是挪威。\n", + "- 队列研究对象的性别分布为全体。\n", + "- 队列收集结束时，研究对象年龄分布均值约为68.8岁，标准差为11.0年。\n", + "- 队列研究时间线为1998年开始收集信息/建立队列，2007年12月31日结束。\n", + "- 队列结束时实际参与研究人数是6261人。\n" + ] + } + ], + "source": [ + "print(rag.query(prompt1, param=QueryParam(mode=\"hybrid\")))" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "fef9d06983da47af", + "metadata": {}, + "outputs": [], + "source": [] + } + ], + "metadata": { + "kernelspec": { + "display_name": "Python 3", + "language": "python", + "name": "python3" + }, + "language_info": { + "codemirror_mode": { + "name": "ipython", + "version": 2 + }, + "file_extension": ".py", + "mimetype": "text/x-python", + "name": "python", + "nbconvert_exporter": "python", + "pygments_lexer": "ipython2", + "version": "2.7.6" + } + }, + "nbformat": 4, + "nbformat_minor": 5 +}