Add mlp_only_layers, decoder_sparse_step, norm_topk_prob; set sliding_window=0 for vLLM 0.21.0 native Laguna
Browse files- config.json +10 -3
config.json
CHANGED
|
@@ -33,7 +33,6 @@
|
|
| 33 |
"use_cache": true,
|
| 34 |
"torch_dtype": "bfloat16",
|
| 35 |
"gating": "per-element",
|
| 36 |
-
"sliding_window": null,
|
| 37 |
"rope_parameters": {
|
| 38 |
"full_attention": {
|
| 39 |
"rope_theta": 500000.0,
|
|
@@ -515,5 +514,13 @@
|
|
| 515 |
"sparsity_config": {},
|
| 516 |
"transform_config": {},
|
| 517 |
"version": "0.11.0"
|
| 518 |
-
}
|
| 519 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 33 |
"use_cache": true,
|
| 34 |
"torch_dtype": "bfloat16",
|
| 35 |
"gating": "per-element",
|
|
|
|
| 36 |
"rope_parameters": {
|
| 37 |
"full_attention": {
|
| 38 |
"rope_theta": 500000.0,
|
|
|
|
| 514 |
"sparsity_config": {},
|
| 515 |
"transform_config": {},
|
| 516 |
"version": "0.11.0"
|
| 517 |
+
},
|
| 518 |
+
"norm_topk_prob": true,
|
| 519 |
+
"mlp_only_layers": [
|
| 520 |
+
0,
|
| 521 |
+
1,
|
| 522 |
+
2
|
| 523 |
+
],
|
| 524 |
+
"decoder_sparse_step": 1,
|
| 525 |
+
"sliding_window": 0
|
| 526 |
+
}
|