From d0a4ef252ee480579777cdfda164651e0118bdf6 Mon Sep 17 00:00:00 2001 From: david Date: Tue, 10 Dec 2024 09:00:22 +0800 Subject: [PATCH 01/11] fix: rare embedding issue. --- lightrag/storage.py | 14 ++++++++++---- 1 file changed, 10 insertions(+), 4 deletions(-) diff --git a/lightrag/storage.py b/lightrag/storage.py index 007d6534..4c043893 100644 --- a/lightrag/storage.py +++ b/lightrag/storage.py @@ -107,10 +107,16 @@ async def upsert(self, data: dict[str, dict]): embeddings = await f embeddings_list.append(embeddings) embeddings = np.concatenate(embeddings_list) - for i, d in enumerate(list_data): - d["__vector__"] = embeddings[i] - results = self._client.upsert(datas=list_data) - return results + if len(embeddings) == len(list_data): + for i, d in enumerate(list_data): + d["__vector__"] = embeddings[i] + results = self._client.upsert(datas=list_data) + return results + else: + # sometimes the embedding is not returned correctly. just log it. + logger.error( + f"embedding is not 1-1 with data, {len(embeddings)} != {len(list_data)}" + ) async def query(self, query: str, top_k=5): embedding = await self.embedding_func([query]) From f6eeedb050a9bdf7d2c598634c76fa152de8e69b Mon Sep 17 00:00:00 2001 From: david Date: Tue, 10 Dec 2024 09:01:21 +0800 Subject: [PATCH 02/11] add concurrent embedding limit --- lightrag/utils.py | 21 ++++++++++++++++++++- 1 file changed, 20 insertions(+), 1 deletion(-) diff --git a/lightrag/utils.py b/lightrag/utils.py index 0220af06..bdb47592 100644 --- a/lightrag/utils.py +++ b/lightrag/utils.py @@ -17,6 +17,17 @@ from lightrag.prompt import PROMPTS + +class UnlimitedSemaphore: + """A context manager that allows unlimited access.""" + + async def __aenter__(self): + pass + + async def __aexit__(self, exc_type, exc, tb): + pass + + ENCODER = None logger = logging.getLogger("lightrag") @@ -42,9 +53,17 @@ class EmbeddingFunc: embedding_dim: int max_token_size: int func: callable + concurrent_limit: int = 16 + + def __post_init__(self): + if self.concurrent_limit != 0: + self._semaphore = asyncio.Semaphore(self.concurrent_limit) + else: + self._semaphore = UnlimitedSemaphore() async def __call__(self, *args, **kwargs) -> np.ndarray: - return await self.func(*args, **kwargs) + async with self._semaphore: + return await self.func(*args, **kwargs) def locate_json_string_body_from_string(content: str) -> Union[str, None]: From 21a3992e397e35449769f9d878e74af6f7a581ab Mon Sep 17 00:00:00 2001 From: david Date: Tue, 10 Dec 2024 09:52:27 +0800 Subject: [PATCH 03/11] fix extra keyword_extraction. --- lightrag/llm.py | 1 + 1 file changed, 1 insertion(+) diff --git a/lightrag/llm.py b/lightrag/llm.py index d725ea85..dbf32844 100644 --- a/lightrag/llm.py +++ b/lightrag/llm.py @@ -63,6 +63,7 @@ async def openai_complete_if_cache( AsyncOpenAI() if base_url is None else AsyncOpenAI(base_url=base_url) ) kwargs.pop("hashing_kv", None) + kwargs.pop("keyword_extraction", None) messages = [] if system_prompt: messages.append({"role": "system", "content": system_prompt}) From 58c0f943464852d7f88c6a766e981c7563d13b7f Mon Sep 17 00:00:00 2001 From: Magic_yuan <317617749@qq.com> Date: Tue, 10 Dec 2024 14:13:11 +0800 Subject: [PATCH 04/11] =?UTF-8?q?fix(lightrag):=20=E4=BF=AE=E5=A4=8D?= =?UTF-8?q?=E5=8F=AA=E6=9C=89=E5=AE=9E=E4=BD=93=E6=B2=A1=E6=9C=89=E5=85=B3?= =?UTF-8?q?=E7=B3=BB=E7=9A=84chunk=E5=A4=84=E7=90=86=E9=80=BB=E8=BE=91=20-?= =?UTF-8?q?=20=E5=8F=AA=E6=9C=89=E5=AE=9E=E4=BD=93=E6=B2=A1=E6=9C=89?= =?UTF-8?q?=E5=85=B3=E7=B3=BB=E6=97=B6=EF=BC=8C=E7=BB=A7=E7=BB=AD=E5=A4=84?= =?UTF-8?q?=E7=90=86=EF=BC=8C=E8=80=8C=E4=B8=8D=E6=98=AF=E7=9B=B4=E6=8E=A5?= =?UTF-8?q?return=20-=20=E5=BD=93=E5=8F=AA=E6=9C=89=E5=AE=9E=E4=BD=93?= =?UTF-8?q?=E8=80=8C=E6=B2=A1=E6=9C=89=E5=85=B3=E7=B3=BB=E7=9A=84=E5=9B=BE?= =?UTF-8?q?=E7=89=87=E5=9C=A8=E9=AB=98=E9=98=B6=E6=9F=A5=E8=AF=A2=E5=85=B3?= =?UTF-8?q?=E7=B3=BB=E6=97=B6=E4=BC=9A=E8=BF=94=E5=9B=9E=E7=A9=BA=EF=BC=8C?= =?UTF-8?q?=E8=BF=99=E9=87=8C=E4=BC=98=E5=8C=96=E8=BF=94=E5=9B=9E=EF=BC=8C?= =?UTF-8?q?=E5=BD=93=E6=B2=A1=E6=9C=89=E5=85=B3=E7=B3=BB=E6=97=B6=E9=99=8D?= =?UTF-8?q?=E7=BA=A7=E4=B8=BAlocal=E6=9F=A5=E8=AF=A2?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- lightrag/operate.py | 21 +++++++++++++++------ 1 file changed, 15 insertions(+), 6 deletions(-) diff --git a/lightrag/operate.py b/lightrag/operate.py index 468f4b2f..ec55694d 100644 --- a/lightrag/operate.py +++ b/lightrag/operate.py @@ -412,15 +412,17 @@ async def _process_single_content(chunk_key_dp: tuple[str, TextChunkSchema]): ): all_relationships_data.append(await result) - if not len(all_entities_data): - logger.warning("Didn't extract any entities, maybe your LLM is not working") - return None - if not len(all_relationships_data): + if not len(all_entities_data) and not len(all_relationships_data): logger.warning( - "Didn't extract any relationships, maybe your LLM is not working" + "Didn't extract any entities or relationships, maybe your LLM is not working" ) return None + if not len(all_entities_data): + logger.warning("Didn't extract any entities") + if not len(all_relationships_data): + logger.warning("Didn't extract any relationships") + if entity_vdb is not None: data_for_vdb = { compute_mdhash_id(dp["entity_name"], prefix="ent-"): { @@ -630,6 +632,13 @@ async def _build_query_context( text_chunks_db, query_param, ) + if ( + 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" if query_param.mode == "hybrid": entities_context, relations_context, text_units_context = combine_contexts( [hl_entities_context, ll_entities_context], @@ -865,7 +874,7 @@ async def _get_edge_data( results = await relationships_vdb.query(keywords, top_k=query_param.top_k) if not len(results): - return None + return "", "", "" edge_datas = await asyncio.gather( *[knowledge_graph_inst.get_edge(r["src_id"], r["tgt_id"]) for r in results] From 316c4df949eda0590e0e33eae07e6cbd386edfc3 Mon Sep 17 00:00:00 2001 From: Magic_yuan <317617749@qq.com> Date: Tue, 10 Dec 2024 14:15:43 +0800 Subject: [PATCH 05/11] =?UTF-8?q?=E6=9B=B4=E6=96=B0=E6=97=A5=E5=BF=97?= =?UTF-8?q?=E6=8F=8F=E8=BF=B0?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- lightrag/operate.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/lightrag/operate.py b/lightrag/operate.py index ec55694d..bc5a9b13 100644 --- a/lightrag/operate.py +++ b/lightrag/operate.py @@ -414,7 +414,7 @@ async def _process_single_content(chunk_key_dp: tuple[str, TextChunkSchema]): if not len(all_entities_data) and not len(all_relationships_data): logger.warning( - "Didn't extract any entities or relationships, maybe your LLM is not working" + "Didn't extract any entities and relationships, maybe your LLM is not working" ) return None From 288d4b8355eb3fd1fdfaf165f1380501b58aaf67 Mon Sep 17 00:00:00 2001 From: david Date: Tue, 10 Dec 2024 17:16:21 +0800 Subject: [PATCH 06/11] fix lazy import --- lightrag/lightrag.py | 19 +++++++++++++------ 1 file changed, 13 insertions(+), 6 deletions(-) diff --git a/lightrag/lightrag.py b/lightrag/lightrag.py index 833926e5..3a4276cb 100644 --- a/lightrag/lightrag.py +++ b/lightrag/lightrag.py @@ -48,18 +48,25 @@ def lazy_external_import(module_name: str, class_name: str): - """Lazily import an external module and return a class from it.""" + """Lazily import a class from an external module based on the package of the caller.""" - def import_class(): + # Get the caller's module and package + import inspect + + caller_frame = inspect.currentframe().f_back + module = inspect.getmodule(caller_frame) + package = module.__package__ if module else None + + def import_class(*args, **kwargs): import importlib # Import the module using importlib - module = importlib.import_module(module_name) + module = importlib.import_module(module_name, package=package) - # Get the class from the module - return getattr(module, class_name) + # Get the class from the module and instantiate it + cls = getattr(module, class_name) + return cls(*args, **kwargs) - # Return the import_class function itself, not its result return import_class From 0a41cc8a9aa33bb3d91f8ea6290e3423721eda9f Mon Sep 17 00:00:00 2001 From: Magic_yuan <317617749@qq.com> Date: Wed, 11 Dec 2024 12:45:10 +0800 Subject: [PATCH 07/11] =?UTF-8?q?feat(llm,=20prompt):=E5=A2=9E=E5=8A=A0?= =?UTF-8?q?=E6=97=A5=E5=BF=97=E8=BE=93=E5=87=BA=E5=B9=B6=E6=89=A9=E5=B1=95?= =?UTF-8?q?=E5=AE=9E=E4=BD=93=E7=B1=BB=E5=9E=8B?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit - 在 llm.py 中添加了日志输出，用于调试和记录 LLM 查询输入 - 在 prompt.py 中增加了 "category" 实体类型，扩展了实体提取的范围 --- lightrag/llm.py | 7 ++++++- lightrag/prompt.py | 2 +- 2 files changed, 7 insertions(+), 2 deletions(-) diff --git a/lightrag/llm.py b/lightrag/llm.py index d725ea85..e0277248 100644 --- a/lightrag/llm.py +++ b/lightrag/llm.py @@ -29,7 +29,7 @@ from .utils import ( wrap_embedding_func_with_attrs, locate_json_string_body_from_string, - safe_unicode_decode, + safe_unicode_decode, logger, ) import sys @@ -69,6 +69,11 @@ async def openai_complete_if_cache( messages.extend(history_messages) messages.append({"role": "user", "content": prompt}) + # 添加日志输出 + logger.debug("===== Query Input to LLM =====") + logger.debug(f"Query: {prompt}") + logger.debug(f"System prompt: {system_prompt}") + logger.debug("Full context:") if "response_format" in kwargs: response = await openai_async_client.beta.chat.completions.parse( model=model, messages=messages, **kwargs diff --git a/lightrag/prompt.py b/lightrag/prompt.py index b62f02b5..d5674f15 100644 --- a/lightrag/prompt.py +++ b/lightrag/prompt.py @@ -8,7 +8,7 @@ PROMPTS["DEFAULT_COMPLETION_DELIMITER"] = "<|COMPLETE|>" PROMPTS["process_tickers"] = ["⠋", "⠙", "⠹", "⠸", "⠼", "⠴", "⠦", "⠧", "⠇", "⠏"] -PROMPTS["DEFAULT_ENTITY_TYPES"] = ["organization", "person", "geo", "event"] +PROMPTS["DEFAULT_ENTITY_TYPES"] = ["organization", "person", "geo", "event", "category"] PROMPTS["entity_extraction"] = """-Goal- Given a text document that is potentially relevant to this activity and a list of entity types, identify all entities of those types from the text and all relationships among the identified entities. From a09e1ba190c155d64ebcbfd98498621068e111de Mon Sep 17 00:00:00 2001 From: Magic_yuan <317617749@qq.com> Date: Wed, 11 Dec 2024 12:57:58 +0800 Subject: [PATCH 08/11] =?UTF-8?q?refactor(prompt):=20=E4=BC=98=E5=8C=96?= =?UTF-8?q?=E6=8F=90=E7=A4=BA=E6=A8=A1=E6=9D=BF=E4=BB=A5=E6=8F=90=E9=AB=98?= =?UTF-8?q?=E7=9B=B8=E4=BC=BC=E5=BA=A6=E8=AF=84=E4=BC=B0=E7=9A=84=E5=87=86?= =?UTF-8?q?=E7=A1=AE=E6=80=A7-=20=E6=98=8E=E7=A1=AE=E4=BA=86=E7=9B=B8?= =?UTF-8?q?=E4=BC=BC=E5=BA=A6=E8=AF=84=E5=88=86=E7=9A=84=E8=AF=84=E5=88=A4?= =?UTF-8?q?=E6=A0=87=E5=87=86=EF=BC=8C=E5=8C=85=E6=8B=AC=E4=B8=8D=E5=90=8C?= =?UTF-8?q?=E6=83=85=E5=86=B5=E4=B8=8B=E7=9A=84=E8=AF=84=E5=88=86=E4=BE=9D?= =?UTF-8?q?=E6=8D=AE=20-=20=E7=AE=80=E5=8C=96=E4=BA=86=E8=AF=84=E5=88=86?= =?UTF-8?q?=E6=B5=81=E7=A8=8B=EF=BC=8C=E8=A6=81=E6=B1=82=E7=9B=B4=E6=8E=A5?= =?UTF-8?q?=E8=BF=94=E5=9B=9E=E6=95=B0=E5=AD=97=20-=20=E6=9C=9F=E6=9C=9B?= =?UTF-8?q?=E9=80=9A=E8=BF=87=E8=BF=99=E4=BA=9B=E6=94=B9=E5=8A=A8=E6=8F=90?= =?UTF-8?q?=E9=AB=98=E8=AF=84=E4=BC=B0=E7=9A=84=E5=87=86=E7=A1=AE=E6=80=A7?= =?UTF-8?q?=E5=92=8C=E4=B8=80=E8=87=B4=E6=80=A7?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- lightrag/prompt.py | 15 ++++++++++----- 1 file changed, 10 insertions(+), 5 deletions(-) diff --git a/lightrag/prompt.py b/lightrag/prompt.py index d5674f15..9d9e6034 100644 --- a/lightrag/prompt.py +++ b/lightrag/prompt.py @@ -268,14 +268,19 @@ Question 1: {original_prompt} Question 2: {cached_prompt} -Please evaluate: +Please evaluate the following two points and provide a similarity score between 0 and 1 directly: 1. Whether these two questions are semantically similar 2. Whether the answer to Question 2 can be used to answer Question 1 - -Please provide a similarity score between 0 and 1, where: -0: Completely unrelated or answer cannot be reused +Similarity score criteria: +0: Completely unrelated or answer cannot be reused, including but not limited to: + - The questions have different topics + - The locations mentioned in the questions are different + - The times mentioned in the questions are different + - The specific individuals mentioned in the questions are different + - The specific events mentioned in the questions are different + - The background information in the questions is different + - The key conditions in the questions are different 1: Identical and answer can be directly reused 0.5: Partially related and answer needs modification to be used - Return only a number between 0-1, without any additional content. """ From b89041b5b38c5d4b2185fe9bc7b84d557d6ac981 Mon Sep 17 00:00:00 2001 From: Magic_yuan <317617749@qq.com> Date: Wed, 11 Dec 2024 13:53:05 +0800 Subject: [PATCH 09/11] =?UTF-8?q?feat(operate):=20=E6=B7=BB=E5=8A=A0?= =?UTF-8?q?=E5=AE=9E=E4=BD=93=E7=B1=BB=E5=9E=8B=E9=85=8D=E7=BD=AE=E5=B9=B6?= =?UTF-8?q?=E4=BC=98=E5=8C=96=E6=8F=90=E7=A4=BA=E7=94=9F=E6=88=90?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit - 在全局配置中添加 entity_types 参数，用于自定义实体类型 - 在生成实体提取和关系提取的提示时，使用配置的实体类型替代默认值 - 优化了提示生成逻辑，提高了代码的可配置性和灵活性 --- lightrag/operate.py | 7 +++++-- 1 file changed, 5 insertions(+), 2 deletions(-) diff --git a/lightrag/operate.py b/lightrag/operate.py index bc5a9b13..8b8ad85b 100644 --- a/lightrag/operate.py +++ b/lightrag/operate.py @@ -260,6 +260,9 @@ async def extract_entities( language = global_config["addon_params"].get( "language", PROMPTS["DEFAULT_LANGUAGE"] ) + entity_types = global_config["addon_params"].get( + "entity_types", PROMPTS["DEFAULT_ENTITY_TYPES"] + ) example_number = global_config["addon_params"].get("example_number", None) if example_number and example_number < len(PROMPTS["entity_extraction_examples"]): examples = "\n".join( @@ -272,7 +275,7 @@ async def extract_entities( tuple_delimiter=PROMPTS["DEFAULT_TUPLE_DELIMITER"], record_delimiter=PROMPTS["DEFAULT_RECORD_DELIMITER"], completion_delimiter=PROMPTS["DEFAULT_COMPLETION_DELIMITER"], - entity_types=",".join(PROMPTS["DEFAULT_ENTITY_TYPES"]), + entity_types=",".join(entity_types), language=language, ) # add example's format @@ -283,7 +286,7 @@ async def extract_entities( tuple_delimiter=PROMPTS["DEFAULT_TUPLE_DELIMITER"], record_delimiter=PROMPTS["DEFAULT_RECORD_DELIMITER"], completion_delimiter=PROMPTS["DEFAULT_COMPLETION_DELIMITER"], - entity_types=",".join(PROMPTS["DEFAULT_ENTITY_TYPES"]), + entity_types=",".join(entity_types), examples=examples, language=language, ) From 9a2afc9484d9a7db93bd2b946a6a963672c327d1 Mon Sep 17 00:00:00 2001 From: Magic_yuan <317617749@qq.com> Date: Wed, 11 Dec 2024 14:06:55 +0800 Subject: [PATCH 10/11] =?UTF-8?q?style(lightrag):=20=E8=B0=83=E6=95=B4?= =?UTF-8?q?=E4=BB=A3=E7=A0=81=E6=A0=BC=E5=BC=8F?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- lightrag/llm.py | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/lightrag/llm.py b/lightrag/llm.py index e0277248..f3fed23f 100644 --- a/lightrag/llm.py +++ b/lightrag/llm.py @@ -29,7 +29,8 @@ from .utils import ( wrap_embedding_func_with_attrs, locate_json_string_body_from_string, - safe_unicode_decode, logger, + safe_unicode_decode, + logger, ) import sys From b63c6155ee0d2e5d8504c1c723b86cec342ae7a8 Mon Sep 17 00:00:00 2001 From: Magic_yuan <317617749@qq.com> Date: Wed, 11 Dec 2024 14:10:27 +0800 Subject: [PATCH 11/11] =?UTF-8?q?style(lightrag):=20=E8=B0=83=E6=95=B4Read?= =?UTF-8?q?Me,=E5=8A=A0=E5=85=A5=E8=87=AA=E5=AE=9A=E4=B9=89=E5=AE=9E?= =?UTF-8?q?=E4=BD=93=E7=B1=BB=E5=9E=8B=E5=8F=82=E6=95=B0=E9=85=8D=E7=BD=AE?= =?UTF-8?q?=E7=A4=BA=E4=BE=8B?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- README.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/README.md b/README.md index a1454792..a24c9b72 100644 --- a/README.md +++ b/README.md @@ -594,7 +594,7 @@ if __name__ == "__main__": | **llm\_model\_kwargs** | `dict` | Additional parameters for LLM generation | | | **vector\_db\_storage\_cls\_kwargs** | `dict` | Additional parameters for vector database (currently not used) | | | **enable\_llm\_cache** | `bool` | If `TRUE`, stores LLM results in cache; repeated prompts return cached responses | `TRUE` | -| **addon\_params** | `dict` | Additional parameters, e.g., `{"example_number": 1, "language": "Simplified Chinese"}`: sets example limit and output language | `example_number: all examples, language: English` | +| **addon\_params** | `dict` | Additional parameters, e.g., `{"example_number": 1, "language": "Simplified Chinese", "entity_types": ["organization", "person", "geo", "event"]}`: sets example limit and output language | `example_number: all examples, language: English` | | **convert\_response\_to\_json\_func** | `callable` | Not used | `convert_response_to_json` | | **embedding\_cache\_config** | `dict` | Configuration for question-answer caching. Contains three parameters:
- `enabled`: Boolean value to enable/disable cache lookup functionality. When enabled, the system will check cached responses before generating new answers.
- `similarity_threshold`: Float value (0-1), similarity threshold. When a new question's similarity with a cached question exceeds this threshold, the cached answer will be returned directly without calling the LLM.
- `use_llm_check`: Boolean value to enable/disable LLM similarity verification. When enabled, LLM will be used as a secondary check to verify the similarity between questions before returning cached answers. | Default: `{"enabled": False, "similarity_threshold": 0.95, "use_llm_check": False}` |