SanjiWatsuki
commited on
Commit
•
c69bcea
1
Parent(s):
61575c1
Update README.md
Browse files
README.md
CHANGED
@@ -81,6 +81,8 @@ parameters:
|
|
81 |
dtype: bfloat16
|
82 |
```
|
83 |
|
|
|
|
|
84 |
### Prompt Template (Alpaca)
|
85 |
|
86 |
```
|
@@ -123,3 +125,5 @@ I found that this model **performed worse** with the xDAN prompt format so, desp
|
|
123 |
| wizardlm-30b | 2 | 6.887500 | 30b
|
124 |
| vicuna-33b-v1.3 | 2 | 6.787500 | 33b
|
125 |
| Llama-2-70b-chat | 2 | 6.725000 | 70b
|
|
|
|
|
|
81 |
dtype: bfloat16
|
82 |
```
|
83 |
|
84 |
+
**There was no additional training, finetuning, or DPO.** This is a straight merger.
|
85 |
+
|
86 |
### Prompt Template (Alpaca)
|
87 |
|
88 |
```
|
|
|
125 |
| wizardlm-30b | 2 | 6.887500 | 30b
|
126 |
| vicuna-33b-v1.3 | 2 | 6.787500 | 33b
|
127 |
| Llama-2-70b-chat | 2 | 6.725000 | 70b
|
128 |
+
|
129 |
+
If you'd like to replicate the MT-Bench run, please ensure that the Alpaca prompt template is applied to the model. I did this by putting "alpaca" in the model path to trigger the `AlpacaAdapter`.
|