numb3r3 qherreros commited on
Commit
050e171
·
verified ·
1 Parent(s): 7fa51ea

add: max_query/doc_length parametrization (#6)

Browse files

- add: max_query/doc_length parametrization (bf1c93a54d0e7666b430d011283feeae5d9a1cf0)
- fix: deactivate automatic ordering (b39421b0c8343e2100221a9ad9b35534f0099c2e)


Co-authored-by: Quentin Herreros <qherreros@users.noreply.huggingface.co>

Files changed (1) hide show
  1. modeling.py +2 -2
modeling.py CHANGED
@@ -200,6 +200,8 @@ class JinaForRanking(modeling_qwen3.Qwen3ForCausalLM):
200
  documents: List[str],
201
  top_n: Optional[int] = None,
202
  return_embeddings: bool = False,
 
 
203
  ) -> List[dict]:
204
  """
205
  Rerank documents by relevance to a query.
@@ -221,8 +223,6 @@ class JinaForRanking(modeling_qwen3.Qwen3ForCausalLM):
221
 
222
  # Derived from model configuration
223
  max_length = self._tokenizer.model_max_length
224
- max_query_length = 512
225
- max_doc_length = 2048
226
 
227
  # Derive block_size from max_length to fit documents efficiently
228
  # Heuristic: allow ~125 docs per batch for typical doc sizes
 
200
  documents: List[str],
201
  top_n: Optional[int] = None,
202
  return_embeddings: bool = False,
203
+ max_doc_length: int = 2048,
204
+ max_query_length: int = 512,
205
  ) -> List[dict]:
206
  """
207
  Rerank documents by relevance to a query.
 
223
 
224
  # Derived from model configuration
225
  max_length = self._tokenizer.model_max_length
 
 
226
 
227
  # Derive block_size from max_length to fit documents efficiently
228
  # Heuristic: allow ~125 docs per batch for typical doc sizes