TimeRobber commited on
Commit
4e45263
·
1 Parent(s): ee84732

Update config.json

Browse files

This model uses sliding window attention I believe. cc

@LucileSaulnier

Files changed (1) hide show
  1. config.json +1 -0
config.json CHANGED
@@ -15,6 +15,7 @@
15
  "num_key_value_heads": 8,
16
  "rms_norm_eps": 1e-05,
17
  "rope_theta": 10000.0,
 
18
  "tie_word_embeddings": false,
19
  "torch_dtype": "bfloat16",
20
  "transformers_version": "4.34.0.dev0",
 
15
  "num_key_value_heads": 8,
16
  "rms_norm_eps": 1e-05,
17
  "rope_theta": 10000.0,
18
+ "sliding_window": 4096,
19
  "tie_word_embeddings": false,
20
  "torch_dtype": "bfloat16",
21
  "transformers_version": "4.34.0.dev0",