Update prompting suggestions
Browse files
README.md
CHANGED
@@ -11,6 +11,7 @@ datasets:
|
|
11 |
- argilla/ultrafeedback-binarized-preferences-cleaned
|
12 |
widget:
|
13 |
- example_title: "Example interaction"
|
|
|
14 |
text: "Why is the sky blue?"
|
15 |
model-index:
|
16 |
- name: phi-2-orange-v2
|
@@ -133,7 +134,7 @@ within Hugging Face's Transformers library (without the need for trust remote co
|
|
133 |
Phi-2 Orange v2 uses ChatML as the prompt format, with or without the system instruction.
|
134 |
(Update 12th March 2024: fixed eos_token issue)
|
135 |
|
136 |
-
|
137 |
|
138 |
```
|
139 |
<|im_start|>system
|
@@ -144,13 +145,14 @@ Write a function to calculate the Fibonacci sequence<|im_end|>
|
|
144 |
|
145 |
```
|
146 |
|
147 |
-
|
148 |
|
149 |
```
|
|
|
|
|
150 |
<|im_start|>user
|
151 |
-
|
152 |
<|im_start|>assistant
|
153 |
-
|
154 |
```
|
155 |
|
156 |
# Evaluations
|
|
|
11 |
- argilla/ultrafeedback-binarized-preferences-cleaned
|
12 |
widget:
|
13 |
- example_title: "Example interaction"
|
14 |
+
context: "You are a helpful assistant. Be short and direct in your answers."
|
15 |
text: "Why is the sky blue?"
|
16 |
model-index:
|
17 |
- name: phi-2-orange-v2
|
|
|
134 |
Phi-2 Orange v2 uses ChatML as the prompt format, with or without the system instruction.
|
135 |
(Update 12th March 2024: fixed eos_token issue)
|
136 |
|
137 |
+
It's recommended to always prompt with a system instruction (use whatever system prompt you like):
|
138 |
|
139 |
```
|
140 |
<|im_start|>system
|
|
|
145 |
|
146 |
```
|
147 |
|
148 |
+
For example, if you find the model's output to be overly verbose, instruct it to be short and concise:
|
149 |
|
150 |
```
|
151 |
+
<|im_start|>system
|
152 |
+
You are a helpful assistant. Be short and direct in your answers.<|im_end|>
|
153 |
<|im_start|>user
|
154 |
+
Was Tom Hanks in the movie Forrest Gump? If so, who did he play and give details of the plot.<|im_end|>
|
155 |
<|im_start|>assistant
|
|
|
156 |
```
|
157 |
|
158 |
# Evaluations
|