Update README.md
Browse files
README.md
CHANGED
@@ -229,7 +229,7 @@ print(tokenizer.apply_chat_template(chat, tokenize=False))
|
|
229 |
```
|
230 |
</details>
|
231 |
|
232 |
-
##
|
233 |
|
234 |
<details>
|
235 |
<summary>
|
@@ -624,4 +624,20 @@ print(tokenizer.apply_chat_template(chat, tokenize=False))
|
|
624 |
```
|
625 |
|
626 |
In other words, write the first chapter, then use a summarization prompt for it, then include the summary in the next chapter's prompt.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
627 |
</details>
|
|
|
229 |
```
|
230 |
</details>
|
231 |
|
232 |
+
## Prompting strategies
|
233 |
|
234 |
<details>
|
235 |
<summary>
|
|
|
624 |
```
|
625 |
|
626 |
In other words, write the first chapter, then use a summarization prompt for it, then include the summary in the next chapter's prompt.
|
627 |
+
</details>
|
628 |
+
|
629 |
+
<details>
|
630 |
+
<summary>
|
631 |
+
<b>Boolean questions</b>
|
632 |
+
<br>
|
633 |
+
For content filtering and other use-cases which only require a true/false response.
|
634 |
+
</summary>
|
635 |
+
|
636 |
+
The prompts in the fine-tuning dataset are formatted as follows:
|
637 |
+
|
638 |
+
```text
|
639 |
+
True or false - {statement}
|
640 |
+
```
|
641 |
+
|
642 |
+
The model will then, theoretically, respond with only a single word.
|
643 |
</details>
|