Chosen DPO model, trained on 40k examples for 3 epochs with default parameters (see train_dpo.py).
Browse files- adapter_config.json +4 -4
adapter_config.json
CHANGED
|
@@ -20,12 +20,12 @@
|
|
| 20 |
"rank_pattern": {},
|
| 21 |
"revision": null,
|
| 22 |
"target_modules": [
|
| 23 |
-
"
|
| 24 |
-
"fc1",
|
| 25 |
-
"k_proj",
|
| 26 |
"v_proj",
|
|
|
|
| 27 |
"q_proj",
|
| 28 |
-
"
|
|
|
|
| 29 |
],
|
| 30 |
"task_type": "CAUSAL_LM",
|
| 31 |
"use_dora": false,
|
|
|
|
| 20 |
"rank_pattern": {},
|
| 21 |
"revision": null,
|
| 22 |
"target_modules": [
|
| 23 |
+
"dense",
|
|
|
|
|
|
|
| 24 |
"v_proj",
|
| 25 |
+
"fc2",
|
| 26 |
"q_proj",
|
| 27 |
+
"k_proj",
|
| 28 |
+
"fc1"
|
| 29 |
],
|
| 30 |
"task_type": "CAUSAL_LM",
|
| 31 |
"use_dora": false,
|