BramVanroy commited on
Commit
e253f4f
·
verified ·
1 Parent(s): 5990844

Fix word_embedding_dimension in 1_Pooling/config.json

Browse files

1_Pooling/config.json declares word_embedding_dimension: 1024, but this model's actual hidden_size (per config.json) is 2048. This mismatch causes sentence-transformers' MSELoss (which sizes its projection layer via model.get_embedding_dimension(), trusting this file) to build a Linear(1024, ...) layer instead of Linear(2048, ...), raising a shape mismatch at runtime when real 2048-dim embeddings are passed in. It also affects any other pooling-dimension-derived Matryoshka dimension defaults. This PR corrects word_embedding_dimension to 2048 to match config.json's hidden_size; all other pooling settings are unchanged.

Files changed (1) hide show
  1. 1_Pooling/config.json +9 -9
1_Pooling/config.json CHANGED
@@ -1,10 +1,10 @@
1
  {
2
- "word_embedding_dimension": 1024,
3
- "pooling_mode_cls_token": false,
4
- "pooling_mode_mean_tokens": false,
5
- "pooling_mode_max_tokens": false,
6
- "pooling_mode_mean_sqrt_len_tokens": false,
7
- "pooling_mode_weightedmean_tokens": false,
8
- "pooling_mode_lasttoken": true,
9
- "include_prompt": true
10
- }
 
1
  {
2
+ "word_embedding_dimension": 2048,
3
+ "pooling_mode_cls_token": false,
4
+ "pooling_mode_mean_tokens": false,
5
+ "pooling_mode_max_tokens": false,
6
+ "pooling_mode_mean_sqrt_len_tokens": false,
7
+ "pooling_mode_weightedmean_tokens": false,
8
+ "pooling_mode_lasttoken": true,
9
+ "include_prompt": true
10
+ }