Upload config.json with huggingface_hub
Browse files- config.json +4 -4
config.json
CHANGED
|
@@ -1,14 +1,14 @@
|
|
| 1 |
{
|
| 2 |
-
"architecture": "
|
| 3 |
"input_dim": 1024,
|
| 4 |
"dict_size": 4096,
|
| 5 |
-
"topk":
|
| 6 |
-
"activation": "
|
| 7 |
"source_model": "llava-hf/llava-1.5-7b-hf",
|
| 8 |
"source_layer": "vision_tower.hidden_states[-2] (Layer 23)",
|
| 9 |
"num_patches_per_image": 576,
|
| 10 |
"training_epochs": 2,
|
| 11 |
"training_lr": 0.0003,
|
| 12 |
-
"l1_coeff":
|
| 13 |
"total_training_vectors": 6336000
|
| 14 |
}
|
|
|
|
| 1 |
{
|
| 2 |
+
"architecture": "TopKSparseAutoencoder",
|
| 3 |
"input_dim": 1024,
|
| 4 |
"dict_size": 4096,
|
| 5 |
+
"topk": 32,
|
| 6 |
+
"activation": "topk",
|
| 7 |
"source_model": "llava-hf/llava-1.5-7b-hf",
|
| 8 |
"source_layer": "vision_tower.hidden_states[-2] (Layer 23)",
|
| 9 |
"num_patches_per_image": 576,
|
| 10 |
"training_epochs": 2,
|
| 11 |
"training_lr": 0.0003,
|
| 12 |
+
"l1_coeff": null,
|
| 13 |
"total_training_vectors": 6336000
|
| 14 |
}
|