Using seq_length Config Rather than max_position_embeddings
Browse filesNeither of these settings are used to run the model, but are referenced by external spaces like the MTEB leaderboard. Since the model doesn't technically have a max_position_embeddings due to the way ALiBi works, its a little more accurate to report a seq_length.
- config.json +1 -1
config.json
CHANGED
|
@@ -1,6 +1,7 @@
|
|
| 1 |
{
|
| 2 |
"_name_or_path": "hum-lodestone-v1",
|
| 3 |
"alibi_starting_size": 4096,
|
|
|
|
| 4 |
"architectures": [
|
| 5 |
"BertModel"
|
| 6 |
],
|
|
@@ -18,7 +19,6 @@
|
|
| 18 |
"initializer_range": 0.02,
|
| 19 |
"intermediate_size": 3072,
|
| 20 |
"layer_norm_eps": 1e-12,
|
| 21 |
-
"max_position_embeddings": 4096,
|
| 22 |
"model_type": "bert",
|
| 23 |
"num_attention_heads": 12,
|
| 24 |
"num_hidden_layers": 12,
|
|
|
|
| 1 |
{
|
| 2 |
"_name_or_path": "hum-lodestone-v1",
|
| 3 |
"alibi_starting_size": 4096,
|
| 4 |
+
"seq_length": 4096,
|
| 5 |
"architectures": [
|
| 6 |
"BertModel"
|
| 7 |
],
|
|
|
|
| 19 |
"initializer_range": 0.02,
|
| 20 |
"intermediate_size": 3072,
|
| 21 |
"layer_norm_eps": 1e-12,
|
|
|
|
| 22 |
"model_type": "bert",
|
| 23 |
"num_attention_heads": 12,
|
| 24 |
"num_hidden_layers": 12,
|