Kevin Hu
commited on
Commit
·
c530565
1
Parent(s):
bb139f2
refine context length (#1813)
Browse files### What problem does this PR solve?
#1594
### Type of change
- [x] Performance Improvement
- graphrag/index.py +1 -1
graphrag/index.py
CHANGED
|
@@ -68,7 +68,7 @@ def build_knowlege_graph_chunks(tenant_id: str, chunks: List[str], callback, ent
|
|
| 68 |
llm_bdl = LLMBundle(tenant_id, LLMType.CHAT)
|
| 69 |
ext = GraphExtractor(llm_bdl)
|
| 70 |
left_token_count = llm_bdl.max_length - ext.prompt_token_count - 1024
|
| 71 |
-
left_token_count = llm_bdl.max_length * 0.
|
| 72 |
|
| 73 |
assert left_token_count > 0, f"The LLM context length({llm_bdl.max_length}) is smaller than prompt({ext.prompt_token_count})"
|
| 74 |
|
|
|
|
| 68 |
llm_bdl = LLMBundle(tenant_id, LLMType.CHAT)
|
| 69 |
ext = GraphExtractor(llm_bdl)
|
| 70 |
left_token_count = llm_bdl.max_length - ext.prompt_token_count - 1024
|
| 71 |
+
left_token_count = max(llm_bdl.max_length * 0.8, left_token_count)
|
| 72 |
|
| 73 |
assert left_token_count > 0, f"The LLM context length({llm_bdl.max_length}) is smaller than prompt({ext.prompt_token_count})"
|
| 74 |
|