gmakstutis/LightRAG
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Improve KG chunk selection documentation and configuration clarity

Browse source
This commit is contained in:
yangdx 2025-08-15 10:09:44 +08:00
parent 3acb32f547
commit 6cab68bb47
2 changed files with 18 additions and 6 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

16
env.example
View file

@ -71,12 +71,20 @@ ENABLE_LLM_CACHE=true
# MAX_RELATION_TOKENS=10000 # MAX_RELATION_TOKENS=10000
### control the maximum tokens send to LLM (include entities, raltions and chunks) ### control the maximum tokens send to LLM (include entities, raltions and chunks)
# MAX_TOTAL_TOKENS=30000 # MAX_TOTAL_TOKENS=30000
### chunk selection strategies for KG: WEIGHT or VECTOR
KG_CHUNK_PICK_METHOD=VECTOR ### maximum number of related chunks per source entity or relation
### maximum number of related chunks per source entity or relation (higher values increase re-ranking time) ### The chunk picker uses this value to determine the total number of chunks selected from KG(knowledge graph)
### Higher values increase re-ranking time
# RELATED_CHUNK_NUMBER=5 # RELATED_CHUNK_NUMBER=5
### Reranker configuration (Set ENABLE_RERANK to true in reranking model is configed) ### chunk selection strategies
### VECTOR: Pick KG chunks by vector similarity, delivered chunks to the LLM aligning more closely with naive retrieval
### WEIGHT: Pick KG chunks by entity and chunk weight, delivered more solely KG related chunks to the LLM
### If reranking is enabled, the impact of chunk selection strategies will be diminished.
# KG_CHUNK_PICK_METHOD=VECTOR
### Reranking configuration
### Reranker Set ENABLE_RERANK to true in reranking model is configed
# ENABLE_RERANK=True # ENABLE_RERANK=True
### Minimum rerank score for document chunk exclusion (set to 0.0 to keep all chunks, 0.6 or above if LLM is not strong enought) ### Minimum rerank score for document chunk exclusion (set to 0.0 to keep all chunks, 0.6 or above if LLM is not strong enought)
# MIN_RERANK_SCORE=0.0 # MIN_RERANK_SCORE=0.0

8
lightrag/operate.py
View file

@ -2779,9 +2779,12 @@ async def _find_related_text_unit_from_entities(
entity_info["sorted_chunks"] = sorted_chunks entity_info["sorted_chunks"] = sorted_chunks
total_entity_chunks += len(sorted_chunks) total_entity_chunks += len(sorted_chunks)
# Step 4: Apply the selected chunk selection algorithm
selected_chunk_ids = [] # Initialize to avoid UnboundLocalError selected_chunk_ids = [] # Initialize to avoid UnboundLocalError
# Step 4: Apply the selected chunk selection algorithm
# Pick by vector similarity:
# The order of text chunks aligns with the naive retrieval's destination.
# When reranking is disabled, the text chunks delivered to the LLM tend to favor naive retrieval.
if kg_chunk_pick_method == "VECTOR" and query and chunks_vdb: if kg_chunk_pick_method == "VECTOR" and query and chunks_vdb:
num_of_chunks = int(max_related_chunks * len(entities_with_chunks) / 2) num_of_chunks = int(max_related_chunks * len(entities_with_chunks) / 2)
@ -2822,7 +2825,8 @@ async def _find_related_text_unit_from_entities(
kg_chunk_pick_method = "WEIGHT" kg_chunk_pick_method = "WEIGHT"
if kg_chunk_pick_method == "WEIGHT": if kg_chunk_pick_method == "WEIGHT":
# Apply linear gradient weighted polling algorithm # Pick by entity and chunk weight:
# When reranking is disabled, delivered more solely KG related chunks to the LLM
selected_chunk_ids = pick_by_weighted_polling( selected_chunk_ids = pick_by_weighted_polling(
entities_with_chunks, max_related_chunks, min_related_chunks=1 entities_with_chunks, max_related_chunks, min_related_chunks=1
) )