johannhartmann commited on
Commit
1dfc973
1 Parent(s): 96daf3e

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +13 -13
README.md CHANGED
@@ -63,25 +63,25 @@ dtype: bfloat16
63
 
64
  ## mt-bench-de
65
 
66
- The results are not bad, but some additional investment into dpo
67
- finetuning would probably help a lot.
68
 
69
  ```json
70
  {
71
- "first_turn": 6.4625,
72
- "second_turn": 5.6375,
73
  "categories": {
74
- "writing": 7.6,
75
- "roleplay": 7.5,
76
- "reasoning": 4.25,
77
- "math": 3.35,
78
- "coding": 3.1,
79
- "extraction": 8.15,
80
- "stem": 6.55,
81
- "humanities": 7.9
82
  },
83
- "average": 6.050000000000001
84
  }
 
85
  ```
86
 
87
  ## 💻 Usage
 
63
 
64
  ## mt-bench-de
65
 
66
+ Using laser and dpo results in pretty good results.
 
67
 
68
  ```json
69
  {
70
+ "first_turn": 7.51875,
71
+ "second_turn": 6.4,
72
  "categories": {
73
+ "writing": 8.425,
74
+ "roleplay": 8.025,
75
+ "reasoning": 5.45,
76
+ "math": 3.2,
77
+ "coding": 4.95,
78
+ "extraction": 7.525,
79
+ "stem": 8.775,
80
+ "humanities": 9.325
81
  },
82
+ "average": 6.959375
83
  }
84
+
85
  ```
86
 
87
  ## 💻 Usage