ArkaAbacus commited on
Commit
0e9c197
1 Parent(s): 5a5a953

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +8 -8
README.md CHANGED
@@ -33,23 +33,23 @@ This model was built using the Smaug recipe for improving performance on real w
33
  ########## First turn ##########
34
  score
35
  model turn
36
- Llama-3-Smaug-8B 1 8.79375
37
- Meta-Llama-3-8B-Instruct 1 8.53125
38
  ########## Second turn ##########
39
  score
40
  model turn
41
- Llama-3-Smaug-8B 2 7.7250
42
- Meta-Llama-3-8B-Instruct 2 7.5500
43
  ########## Average ##########
44
  score
45
  model
46
- Llama-3-Smaug-8B 8.259375
47
- Meta-Llama-3-8B-Instruct 8.040625
48
  ```
49
 
50
  | Model | First turn | Second Turn | Average |
51
  | :---- | ---------: | ----------: | ------: |
52
- | Llama-3-Smaug-8B | 8.79 | 7.73 | 8.26 |
53
- | Llama-3-8B-Instruct | 8.53 | 7.55 | 8.04 |
54
 
55
  This version of Smaug uses new techniques and new data compared to [Smaug-72B](https://huggingface.co/abacusai/Smaug-72B-v0.1), and more information will be released later on. For now, see the previous Smaug paper: https://arxiv.org/abs/2402.13228.
 
33
  ########## First turn ##########
34
  score
35
  model turn
36
+ Llama-3-Smaug-8B 1 8.77500
37
+ Meta-Llama-3-8B-Instruct 1 8.1
38
  ########## Second turn ##########
39
  score
40
  model turn
41
+ Meta-Llama-3-8B-Instruct 2 8.2125
42
+ Llama-3-Smaug-8B 2 7.8875
43
  ########## Average ##########
44
  score
45
  model
46
+ Llama-3-Smaug-8B 8.331250
47
+ Meta-Llama-3-8B-Instruct 8.15625
48
  ```
49
 
50
  | Model | First turn | Second Turn | Average |
51
  | :---- | ---------: | ----------: | ------: |
52
+ | Llama-3-Smaug-8B | 8.78 | 7.89 | 8.33 |
53
+ | Llama-3-8B-Instruct | 8.1 | 8.21 | 8.16 |
54
 
55
  This version of Smaug uses new techniques and new data compared to [Smaug-72B](https://huggingface.co/abacusai/Smaug-72B-v0.1), and more information will be released later on. For now, see the previous Smaug paper: https://arxiv.org/abs/2402.13228.