mirror of
https://github.com/HKUDS/LightRAG.git
synced 2025-12-03 02:16:42 +00:00
feat: optimize entity extraction for smaller LLMs
Simplify entity relationship extraction process to improve compatibility and performance with smaller, less capable language models. Changes: - Remove iterative gleaning loop with LLM-based continuation decisions - Simplify to single gleaning pass when entity_extract_max_gleaning > 0 - Streamline entity extraction prompts with clearer instructions - Add explicit completion delimiter signals in all examples
This commit is contained in:
yangdx
parent
9d81cd724a
commit
c86f863fa4
10
env.example
10
env.example
@ -175,11 +175,9 @@ LLM_BINDING_API_KEY=your_api_key
|
||||
# LLM_BINDING=openai
|
||||
|
||||
### OpenAI Specific Parameters
|
||||
### To mitigate endless output loops and prevent greedy decoding for Qwen3, set the temperature parameter to a value between 0.8 and 1.0
|
||||
# OPENAI_LLM_TEMPERATURE=1.0
|
||||
# OPENAI_LLM_REASONING_EFFORT=low
|
||||
### If the presence penalty still can not stop the model from generates repetitive or unconstrained output
|
||||
# OPENAI_LLM_MAX_COMPLETION_TOKENS=16384
|
||||
### To mitigate endless output loops and prevent greedy decoding for Qwen3, set the temperature and frequency penalty parameter to a highter value
|
||||
# OPENAI_LLM_TEMPERATURE=1.2
|
||||
# OPENAI_FREQUENCY_PENALTY=1.5
|
||||
|
||||
### OpenRouter Specific Parameters
|
||||
# OPENAI_LLM_EXTRA_BODY='{"reasoning": {"enabled": false}}'
|
||||
@ -194,7 +192,7 @@ LLM_BINDING_API_KEY=your_api_key
|
||||
OLLAMA_LLM_NUM_CTX=32768
|
||||
# OLLAMA_LLM_TEMPERATURE=1.0
|
||||
### Stop sequences for Ollama LLM
|
||||
# OLLAMA_LLM_STOP='["</s>", "Assistant:", "\n\n"]'
|
||||
# OLLAMA_LLM_STOP='["</s>", "<|EOT|>"]'
|
||||
### use the following command to see all support options for Ollama LLM
|
||||
### lightrag-server --llm-binding ollama --help
|
||||
|
||||
|
||||
@ -1764,7 +1764,6 @@ async def extract_entities(
|
||||
)
|
||||
|
||||
continue_prompt = PROMPTS["entity_continue_extraction"].format(**context_base)
|
||||
if_loop_prompt = PROMPTS["entity_if_loop_extraction"]
|
||||
|
||||
processed_chunks = 0
|
||||
total_chunks = len(ordered_chunks)
|
||||
@ -1815,7 +1814,7 @@ async def extract_entities(
|
||||
)
|
||||
|
||||
# Process additional gleaning results
|
||||
for now_glean_index in range(entity_extract_max_gleaning):
|
||||
if entity_extract_max_gleaning > 0:
|
||||
glean_result = await use_llm_func_with_cache(
|
||||
continue_prompt,
|
||||
use_llm_func,
|
||||
@ -1852,21 +1851,6 @@ async def extract_entities(
|
||||
maybe_edges[edge_key] = [] # Explicitly create the list
|
||||
maybe_edges[edge_key].extend(edges)
|
||||
|
||||
if now_glean_index == entity_extract_max_gleaning - 1:
|
||||
break
|
||||
|
||||
if_loop_result: str = await use_llm_func_with_cache(
|
||||
if_loop_prompt,
|
||||
use_llm_func,
|
||||
llm_response_cache=llm_response_cache,
|
||||
history_messages=history,
|
||||
cache_type="extract",
|
||||
cache_keys_collector=cache_keys_collector,
|
||||
)
|
||||
if_loop_result = if_loop_result.strip().strip('"').strip("'").lower()
|
||||
if if_loop_result != "yes":
|
||||
break
|
||||
|
||||
# Batch update chunk's llm_cache_list with all collected cache keys
|
||||
if cache_keys_collector and text_chunks_storage:
|
||||
await update_chunk_cache_list(
|
||||
|
||||
@ -6,37 +6,29 @@ PROMPTS: dict[str, Any] = {}
|
||||
|
||||
PROMPTS["DEFAULT_TUPLE_DELIMITER"] = "<|>"
|
||||
PROMPTS["DEFAULT_RECORD_DELIMITER"] = "##"
|
||||
|
||||
# TODO: Deprecated, reserved for compatible with legacy LLM cache
|
||||
PROMPTS["DEFAULT_COMPLETION_DELIMITER"] = "<|COMPLETE|>"
|
||||
|
||||
PROMPTS["DEFAULT_USER_PROMPT"] = "n/a"
|
||||
|
||||
PROMPTS["entity_extraction"] = """---Task---
|
||||
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.
|
||||
Given a text document and a list of entity types, identify all entities of those types and all relationships among the identified entities.
|
||||
|
||||
---Instructions---
|
||||
1. Recognizing definitively conceptualized entities in text. For each identified entity, extract the following information:
|
||||
- entity_name: Name of the entity, use same language as input text. If English, capitalized the name
|
||||
- entity_type: One of the following types: [{entity_types}]. If the entity doesn't clearly fit any category, classify it as "Other".
|
||||
- entity_description: Provide a comprehensive description of the entity's attributes and activities based on the information present in the input text. Do not add external knowledge.
|
||||
|
||||
2. Format each entity as:
|
||||
("entity"{tuple_delimiter}<entity_name>{tuple_delimiter}<entity_type>{tuple_delimiter}<entity_description>)
|
||||
|
||||
- entity_name: Name of the entity, use same language as input text. If English, capitalized the name
|
||||
- entity_type: Categorize the entity using the provided `Entity_types` list. If a suitable category cannot be determined, classify it as "Other".
|
||||
- entity_description: Provide a comprehensive description of the entity's attributes and activities based on the information present in the input text. Do not add external knowledge.
|
||||
2. Format each entity as: ("entity"{tuple_delimiter}<entity_name>{tuple_delimiter}<entity_type>{tuple_delimiter}<entity_description>)
|
||||
3. From the entities identified in step 1, identify all pairs of (source_entity, target_entity) that are directly and clearly related based on the text. Unsubstantiated relationships must be excluded from the output.
|
||||
For each pair of related entities, extract the following information:
|
||||
- source_entity: name of the source entity, as identified in step 1
|
||||
- target_entity: name of the target entity, as identified in step 1
|
||||
- relationship_keywords: one or more high-level key words that summarize the overarching nature of the relationship, focusing on concepts or themes rather than specific details
|
||||
- relationship_description: Explain the nature of the relationship between the source and target entities, providing a clear rationale for their connection
|
||||
|
||||
4. Format each relationship as:
|
||||
("relationship"{tuple_delimiter}<source_entity>{tuple_delimiter}<target_entity>{tuple_delimiter}<relationship_keywords>{tuple_delimiter}<relationship_description>)
|
||||
|
||||
- source_entity: name of the source entity, as identified in step 1
|
||||
- target_entity: name of the target entity, as identified in step 1
|
||||
- relationship_keywords: one or more high-level key words that summarize the overarching nature of the relationship, focusing on concepts or themes rather than specific details
|
||||
- relationship_description: Explain the nature of the relationship between the source and target entities, providing a clear rationale for their connection
|
||||
4. Format each relationship as: ("relationship"{tuple_delimiter}<source_entity>{tuple_delimiter}<target_entity>{tuple_delimiter}<relationship_keywords>{tuple_delimiter}<relationship_description>)
|
||||
5. Use `{tuple_delimiter}` as field delimiter. Use `{record_delimiter}` as the entity or relation list delimiter.
|
||||
|
||||
6. Return identified entities and relationships in {language}.
|
||||
7. Output `{completion_delimiter}` when all the entities and relationships are extracted.
|
||||
|
||||
---Quality Guidelines---
|
||||
- Only extract entities that are clearly defined and meaningful in the context
|
||||
@ -47,7 +39,7 @@ For each pair of related entities, extract the following information:
|
||||
---Examples---
|
||||
{examples}
|
||||
|
||||
---Real Data---
|
||||
---Input---
|
||||
Entity_types: [{entity_types}]
|
||||
Text:
|
||||
```
|
||||
@ -55,12 +47,12 @@ Text:
|
||||
```
|
||||
|
||||
---Output---
|
||||
Output:
|
||||
"""
|
||||
|
||||
PROMPTS["entity_extraction_examples"] = [
|
||||
"""------Example 1------
|
||||
"""[Example 1]
|
||||
|
||||
---Input---
|
||||
Entity_types: [organization,person,equiment,product,technology,location,event,category]
|
||||
Text:
|
||||
```
|
||||
@ -73,7 +65,7 @@ The underlying dismissal earlier seemed to falter, replaced by a glimpse of relu
|
||||
It was a small transformation, barely perceptible, but one that Alex noted with an inward nod. They had all been brought here by different paths
|
||||
```
|
||||
|
||||
Output:
|
||||
---Output---
|
||||
(entity{tuple_delimiter}Alex{tuple_delimiter}person{tuple_delimiter}Alex is a character who experiences frustration and is observant of the dynamics among other characters.){record_delimiter}
|
||||
(entity{tuple_delimiter}Taylor{tuple_delimiter}person{tuple_delimiter}Taylor is portrayed with authoritarian certainty and shows a moment of reverence towards a device, indicating a change in perspective.){record_delimiter}
|
||||
(entity{tuple_delimiter}Jordan{tuple_delimiter}person{tuple_delimiter}Jordan shares a commitment to discovery and has a significant interaction with Taylor regarding a device.){record_delimiter}
|
||||
@ -84,10 +76,12 @@ Output:
|
||||
(relationship{tuple_delimiter}Taylor{tuple_delimiter}Jordan{tuple_delimiter}conflict resolution, mutual respect{tuple_delimiter}Taylor and Jordan interact directly regarding the device, leading to a moment of mutual respect and an uneasy truce.){record_delimiter}
|
||||
(relationship{tuple_delimiter}Jordan{tuple_delimiter}Cruz{tuple_delimiter}ideological conflict, rebellion{tuple_delimiter}Jordan's commitment to discovery is in rebellion against Cruz's vision of control and order.){record_delimiter}
|
||||
(relationship{tuple_delimiter}Taylor{tuple_delimiter}The Device{tuple_delimiter}reverence, technological significance{tuple_delimiter}Taylor shows reverence towards the device, indicating its importance and potential impact.){record_delimiter}
|
||||
{completion_delimiter}
|
||||
|
||||
""",
|
||||
"""------Example 2------
|
||||
"""[Example 2]
|
||||
|
||||
---Input---
|
||||
Entity_types: [organization,person,equiment,product,technology,location,event,category]
|
||||
Text:
|
||||
```
|
||||
@ -100,7 +94,7 @@ Meanwhile, commodity markets reflected a mixed sentiment. Gold futures rose by 1
|
||||
Financial experts are closely watching the Federal Reserve's next move, as speculation grows over potential rate hikes. The upcoming policy announcement is expected to influence investor confidence and overall market stability.
|
||||
```
|
||||
|
||||
Output:
|
||||
---Output---
|
||||
(entity{tuple_delimiter}Global Tech Index{tuple_delimiter}category{tuple_delimiter}The Global Tech Index tracks the performance of major technology stocks and experienced a 3.4% decline today.){record_delimiter}
|
||||
(entity{tuple_delimiter}Nexon Technologies{tuple_delimiter}organization{tuple_delimiter}Nexon Technologies is a tech company that saw its stock decline by 7.8% after disappointing earnings.){record_delimiter}
|
||||
(entity{tuple_delimiter}Omega Energy{tuple_delimiter}organization{tuple_delimiter}Omega Energy is an energy company that gained 2.1% in stock value due to rising oil prices.){record_delimiter}
|
||||
@ -113,17 +107,19 @@ Output:
|
||||
(relationship{tuple_delimiter}Nexon Technologies{tuple_delimiter}Global Tech Index{tuple_delimiter}company impact, index movement{tuple_delimiter}Nexon Technologies' stock decline contributed to the overall drop in the Global Tech Index.){record_delimiter}
|
||||
(relationship{tuple_delimiter}Gold Futures{tuple_delimiter}Market Selloff{tuple_delimiter}market reaction, safe-haven investment{tuple_delimiter}Gold prices rose as investors sought safe-haven assets during the market selloff.){record_delimiter}
|
||||
(relationship{tuple_delimiter}Federal Reserve Policy Announcement{tuple_delimiter}Market Selloff{tuple_delimiter}interest rate impact, financial regulation{tuple_delimiter}Speculation over Federal Reserve policy changes contributed to market volatility and investor selloff.){record_delimiter}
|
||||
{completion_delimiter}
|
||||
|
||||
""",
|
||||
"""------Example 3------
|
||||
"""[Example 3]
|
||||
|
||||
---Input---
|
||||
Entity_types: [organization,person,equiment,product,technology,location,event,category]
|
||||
Text:
|
||||
```
|
||||
At the World Athletics Championship in Tokyo, Noah Carter broke the 100m sprint record using cutting-edge carbon-fiber spikes.
|
||||
```
|
||||
|
||||
Output:
|
||||
---Output---
|
||||
(entity{tuple_delimiter}World Athletics Championship{tuple_delimiter}event{tuple_delimiter}The World Athletics Championship is a global sports competition featuring top athletes in track and field.){record_delimiter}
|
||||
(entity{tuple_delimiter}Tokyo{tuple_delimiter}location{tuple_delimiter}Tokyo is the host city of the World Athletics Championship.){record_delimiter}
|
||||
(entity{tuple_delimiter}Noah Carter{tuple_delimiter}person{tuple_delimiter}Noah Carter is a sprinter who set a new record in the 100m sprint at the World Athletics Championship.){record_delimiter}
|
||||
@ -134,17 +130,19 @@ Output:
|
||||
(relationship{tuple_delimiter}Noah Carter{tuple_delimiter}100m Sprint Record{tuple_delimiter}athlete achievement, record-breaking{tuple_delimiter}Noah Carter set a new 100m sprint record at the championship.){record_delimiter}
|
||||
(relationship{tuple_delimiter}Noah Carter{tuple_delimiter}Carbon-Fiber Spikes{tuple_delimiter}athletic equipment, performance boost{tuple_delimiter}Noah Carter used carbon-fiber spikes to enhance performance during the race.){record_delimiter}
|
||||
(relationship{tuple_delimiter}Noah Carter{tuple_delimiter}World Athletics Championship{tuple_delimiter}athlete participation, competition{tuple_delimiter}Noah Carter is competing at the World Athletics Championship.){record_delimiter}
|
||||
{completion_delimiter}
|
||||
|
||||
""",
|
||||
"""------Example 4------
|
||||
"""[Example 4]
|
||||
|
||||
---Input---
|
||||
Entity_types: [organization,person,equiment,product,technology,location,event,category]
|
||||
Text:
|
||||
```
|
||||
在北京举行的人工智能大会上，腾讯公司的首席技术官张伟发布了最新的大语言模型"腾讯智言"，该模型在自然语言处理方面取得了重大突破。
|
||||
```
|
||||
|
||||
Output:
|
||||
---Output---
|
||||
(entity{tuple_delimiter}人工智能大会{tuple_delimiter}event{tuple_delimiter}人工智能大会是在北京举行的技术会议，专注于人工智能领域的最新发展。){record_delimiter}
|
||||
(entity{tuple_delimiter}北京{tuple_delimiter}location{tuple_delimiter}北京是人工智能大会的举办城市。){record_delimiter}
|
||||
(entity{tuple_delimiter}腾讯公司{tuple_delimiter}organization{tuple_delimiter}腾讯公司是参与人工智能大会的科技企业，发布了新的语言模型产品。){record_delimiter}
|
||||
@ -155,6 +153,7 @@ Output:
|
||||
(relationship{tuple_delimiter}张伟{tuple_delimiter}腾讯公司{tuple_delimiter}雇佣关系, 高管职位{tuple_delimiter}张伟担任腾讯公司的首席技术官。){record_delimiter}
|
||||
(relationship{tuple_delimiter}张伟{tuple_delimiter}腾讯智言{tuple_delimiter}产品发布, 技术展示{tuple_delimiter}张伟在大会上发布了腾讯智言大语言模型。){record_delimiter}
|
||||
(relationship{tuple_delimiter}腾讯智言{tuple_delimiter}自然语言处理技术{tuple_delimiter}技术应用, 突破创新{tuple_delimiter}腾讯智言在自然语言处理技术方面取得了重大突破。){record_delimiter}
|
||||
{completion_delimiter}
|
||||
|
||||
""",
|
||||
]
|
||||
@ -179,54 +178,29 @@ Description List:
|
||||
{description_list}
|
||||
|
||||
---Output---
|
||||
Output:"""
|
||||
|
||||
PROMPTS["entity_continue_extraction"] = """
|
||||
---Task---
|
||||
MANY entities and relationships were missed in the last extraction. Please find only the missing entities and relationships from previous text.
|
||||
|
||||
---Instructions---
|
||||
1. Recognizing definitively conceptualized entities in text. For each identified entity, extract the following information:
|
||||
- entity_name: Name of the entity, use same language as input text. If English, capitalized the name
|
||||
- entity_type: One of the following types: [{entity_types}]. If the entity doesn't clearly fit any category, classify it as "Other".
|
||||
- entity_description: Provide a comprehensive description of the entity's attributes and activities based on the information present in the input text. Do not add external knowledge.
|
||||
|
||||
2. Format each entity as:
|
||||
("entity"{tuple_delimiter}<entity_name>{tuple_delimiter}<entity_type>{tuple_delimiter}<entity_description>)
|
||||
|
||||
3. From the entities identified in step 1, identify all pairs of (source_entity, target_entity) that are directly and clearly related based on the text. Unsubstantiated relationships must be excluded from the output.
|
||||
For each pair of related entities, extract the following information:
|
||||
- source_entity: name of the source entity, as identified in step 1
|
||||
- target_entity: name of the target entity, as identified in step 1
|
||||
- relationship_keywords: one or more high-level key words that summarize the overarching nature of the relationship, focusing on concepts or themes rather than specific details
|
||||
- relationship_description: Explain the nature of the relationship between the source and target entities, providing a clear rationale for their connection
|
||||
|
||||
4. Format each relationship as:
|
||||
("relationship"{tuple_delimiter}<source_entity>{tuple_delimiter}<target_entity>{tuple_delimiter}<relationship_keywords>{tuple_delimiter}<relationship_description>)
|
||||
|
||||
5. Use `{tuple_delimiter}` as field delimiter. Use `{record_delimiter}` as the entity or relation list delimiter.
|
||||
|
||||
6. Return identified entities and relationships in {language}.
|
||||
|
||||
---Quality Guidelines---
|
||||
- Only extract entities that are clearly defined and meaningful in the context
|
||||
- Do not include entities and relations that have been previously extracted
|
||||
- Avoid over-interpretation; stick to what is explicitly stated in the text
|
||||
- Include specific numerical data in entity name when relevant
|
||||
- Ensure entity names are consistent throughout the extraction
|
||||
|
||||
---Output---
|
||||
Output:
|
||||
"""
|
||||
|
||||
PROMPTS["entity_continue_extraction"] = """---Task---
|
||||
Identify any missed entities or relationships in the last extraction task.
|
||||
|
||||
---Instructions---
|
||||
1. Output the entities and realtionships in the same format as previous extraction task.
|
||||
2. Do not include entities and relations that have been previously extracted.
|
||||
3. If the entity doesn't clearly fit in any of`Entity_types` provided, classify it as "Other".
|
||||
4. Return identified entities and relationships in {language}.
|
||||
5. Output `{completion_delimiter}` when all the entities and relationships are extracted.
|
||||
|
||||
---Output---
|
||||
"""
|
||||
|
||||
# TODO: Deprecated
|
||||
PROMPTS["entity_if_loop_extraction"] = """
|
||||
---Goal---'
|
||||
|
||||
It appears some entities may have still been missed.
|
||||
Check if it appears some entities may have still been missed. Output "Yes" if so, otherwise "No".
|
||||
|
||||
---Output---
|
||||
Output:
|
||||
""".strip()
|
||||
Output:"""
|
||||
|
||||
PROMPTS["fail_response"] = (
|
||||
"Sorry, I'm not able to provide an answer to that question.[no-context]"
|
||||
@ -270,7 +244,7 @@ Generate a concise response based on Knowledge Base and follow Response Rules, c
|
||||
- Additional user prompt: {user_prompt}
|
||||
|
||||
---Response---
|
||||
Output:"""
|
||||
"""
|
||||
|
||||
PROMPTS["keywords_extraction"] = """---Role---
|
||||
You are an expert keyword extractor, specializing in analyzing user queries for a Retrieval-Augmented Generation (RAG) system. Your purpose is to identify both high-level and low-level keywords in the user's query that will be used for effective document retrieval.
|
||||
|
||||
Loading…
x
Reference in New Issue
Block a user