Our new X account is live! Follow @wizwand_team for updates
Home
/
Benchmarks
Text Generation on Harry Potter forget data (400 chunks)
Loading...
8.02
BLEU
Target LLM
-0.3208
1.8446
4.01
6.1754
Jun 12, 2024
BLEU
ROUGE-L
Updated 4d ago
Evaluation Results
Method
Method
Links
BLEU
ROUGE-L
Target LLM
Backbone=Mistral-7B-in...
2024.06
8.02
16.98
NPO+GD
Backbone=Mistral-7B-in...
2024.06
0.82
5.76
Before finetune
Backbone=Mistral-7B-in...
2024.06
0.74
8.97
NPO+KL
Backbone=Mistral-7B-in...
2024.06
0.74
6.84
ULD
Backbone=Mistral-7B-in...
2024.06
0.67
4.58
Offset-NPO+KL
Backbone=Mistral-7B-in...
2024.06
0.58
8.55
NPO
Backbone=Mistral-7B-in...
2024.06
0.47
4.31
Offset-DPO+KL
Backbone=Mistral-7B-in...
2024.06
0.45
4.39
DPO+GD
Backbone=Mistral-7B-in...
2024.06
0.38
3.94
DPO
Backbone=Mistral-7B-in...
2024.06
0.35
4.24
DPO+KL
Backbone=Mistral-7B-in...
2024.06
0.35
4.15
GA
Backbone=Mistral-7B-in...
2024.06
0
0
GA+GD
Backbone=Mistral-7B-in...
2024.06
0
0
GA+KL
Backbone=Mistral-7B-in...
2024.06
0
0
Offset-GA+KL
Backbone=Mistral-7B-in...
2024.06
0
0
Feedback
Search any
task
Search any
task