Update README.md
Browse filesPrompt information
README.md
CHANGED
|
@@ -7,4 +7,15 @@ EQ-Bench: 84.89
|
|
| 7 |
GSM8k: 77.18
|
| 8 |
Hellaswag: 87.67
|
| 9 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 10 |
Credit to https://twitter.com/hu_yifei for providing GSM & Hellaswag. It is the first open weight model to dethrone GPT-4 on EQ bench,
|
|
|
|
| 7 |
GSM8k: 77.18
|
| 8 |
Hellaswag: 87.67
|
| 9 |
|
| 10 |
+
Edit: Upon further testing a score of 85.09 was achieved using ChatLM instead of Mistral's prompt.
|
| 11 |
+
|
| 12 |
+
I recommend using the ChatML format instead, I will run more benchmarks. This also fixes the bug with Miqu dequant failing to provide a stop.
|
| 13 |
+
<|im_start|>system
|
| 14 |
+
Provide some context and/or instructions to the model.
|
| 15 |
+
<|im_end|>
|
| 16 |
+
<|im_start|>user
|
| 17 |
+
The user’s message goes here
|
| 18 |
+
<|im_end|>
|
| 19 |
+
<|im_start|>assistant <|im_end|>
|
| 20 |
+
|
| 21 |
Credit to https://twitter.com/hu_yifei for providing GSM & Hellaswag. It is the first open weight model to dethrone GPT-4 on EQ bench,
|