Update LoRA adapter with dataset v2
Browse files- README.md +9 -0
- adapter_config.json +5 -5
- adapter_model.safetensors +1 -1
README.md
CHANGED
|
@@ -38,6 +38,15 @@ while intermediate reasoning (Chain-of-Thought) is masked.
|
|
| 38 |
- Learning rate: 5e-06
|
| 39 |
- LoRA: r=64, alpha=128
|
| 40 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 41 |
## Usage
|
| 42 |
|
| 43 |
```python
|
|
|
|
| 38 |
- Learning rate: 5e-06
|
| 39 |
- LoRA: r=64, alpha=128
|
| 40 |
|
| 41 |
+
### Data Sampling Strategy
|
| 42 |
+
- Upsampling:
|
| 43 |
+
- **yaml**: 1.2x
|
| 44 |
+
- **xml**: 1.2x
|
| 45 |
+
- **toml**: 1.2x
|
| 46 |
+
- **json**: 1.0x
|
| 47 |
+
- **csv**: 1.0x
|
| 48 |
+
(Categories with weights > 1.0 were sampled more frequently to correct specific format weaknesses.)
|
| 49 |
+
|
| 50 |
## Usage
|
| 51 |
|
| 52 |
```python
|
adapter_config.json
CHANGED
|
@@ -33,13 +33,13 @@
|
|
| 33 |
"rank_pattern": {},
|
| 34 |
"revision": null,
|
| 35 |
"target_modules": [
|
| 36 |
-
"k_proj",
|
| 37 |
-
"v_proj",
|
| 38 |
-
"q_proj",
|
| 39 |
-
"down_proj",
|
| 40 |
"o_proj",
|
|
|
|
| 41 |
"gate_proj",
|
| 42 |
-
"
|
|
|
|
|
|
|
|
|
|
| 43 |
],
|
| 44 |
"target_parameters": null,
|
| 45 |
"task_type": "CAUSAL_LM",
|
|
|
|
| 33 |
"rank_pattern": {},
|
| 34 |
"revision": null,
|
| 35 |
"target_modules": [
|
|
|
|
|
|
|
|
|
|
|
|
|
| 36 |
"o_proj",
|
| 37 |
+
"v_proj",
|
| 38 |
"gate_proj",
|
| 39 |
+
"k_proj",
|
| 40 |
+
"up_proj",
|
| 41 |
+
"down_proj",
|
| 42 |
+
"q_proj"
|
| 43 |
],
|
| 44 |
"target_parameters": null,
|
| 45 |
"task_type": "CAUSAL_LM",
|
adapter_model.safetensors
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 528550256
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:1b52683cdb3b12d9cbc644270ece87b040b87712d45a8569ae16985bc8a26b53
|
| 3 |
size 528550256
|