instruction-pretrain
commited on
Commit
•
bbcfb36
1
Parent(s):
db13bd2
Update README.md
Browse files
README.md
CHANGED
@@ -17,15 +17,17 @@ We explore supervised multitask pre-training by proposing ***Instruction Pre-Tra
|
|
17 |
</p>
|
18 |
|
19 |
**************************** **Updates** ****************************
|
|
|
20 |
* 2024/7/15: We scaled up the pre-trained tokens from 100B to 250B, with the number of synthesized instruction-response pairs reaching 500M! Below, we show the performance trend on downstream tasks throughout the pre-training process:
|
21 |
-
<p align='
|
22 |
-
<img src="https://cdn-uploads.huggingface.co/production/uploads/66711d2ee12fa6cc5f5dfc89/0okCfRkC6uALTfuNxt0Fa.png" width="
|
23 |
</p>
|
24 |
* 2024/6/21: Released the [paper](https://huggingface.co/papers/2406.14491), [code](https://github.com/microsoft/LMOps), and [resources](https://huggingface.co/instruction-pretrain)
|
25 |
|
26 |
## Resources
|
27 |
-
**🤗 We share our data and models with example usages, feel free to open any
|
28 |
|
|
|
29 |
- Context-Based Instruction Synthesizer: [instruction-synthesizer](https://huggingface.co/instruction-pretrain/instruction-synthesizer)
|
30 |
- Fine-Tuning Data for the Synthesizer: [ft-instruction-synthesizer-collection](https://huggingface.co/datasets/instruction-pretrain/ft-instruction-synthesizer-collection)
|
31 |
- General Models Pre-Trained from Scratch (on 100B tokes):
|
@@ -82,7 +84,7 @@ Instruction Pre-Training
|
|
82 |
}
|
83 |
```
|
84 |
|
85 |
-
[
|
86 |
```bibtex
|
87 |
@inproceedings{
|
88 |
cheng2024adapting,
|
|
|
17 |
</p>
|
18 |
|
19 |
**************************** **Updates** ****************************
|
20 |
+
* 2024/7/31: Updated pre-training suggestions in the `Advanced Usage` section of [instruction-synthesizer](https://huggingface.co/instruction-pretrain/instruction-synthesizer)
|
21 |
* 2024/7/15: We scaled up the pre-trained tokens from 100B to 250B, with the number of synthesized instruction-response pairs reaching 500M! Below, we show the performance trend on downstream tasks throughout the pre-training process:
|
22 |
+
<p align='left'>
|
23 |
+
<img src="https://cdn-uploads.huggingface.co/production/uploads/66711d2ee12fa6cc5f5dfc89/0okCfRkC6uALTfuNxt0Fa.png" width="500">
|
24 |
</p>
|
25 |
* 2024/6/21: Released the [paper](https://huggingface.co/papers/2406.14491), [code](https://github.com/microsoft/LMOps), and [resources](https://huggingface.co/instruction-pretrain)
|
26 |
|
27 |
## Resources
|
28 |
+
**🤗 We share our data and models with example usages, feel free to open any discussions at [this page](https://huggingface.co/papers/2406.14491)! 🤗**
|
29 |
|
30 |
+
- Thanks to the demo [davanstrien/instruction-synthesizer](https://huggingface.co/spaces/davanstrien/instruction-synthesizer) for implementing our approach
|
31 |
- Context-Based Instruction Synthesizer: [instruction-synthesizer](https://huggingface.co/instruction-pretrain/instruction-synthesizer)
|
32 |
- Fine-Tuning Data for the Synthesizer: [ft-instruction-synthesizer-collection](https://huggingface.co/datasets/instruction-pretrain/ft-instruction-synthesizer-collection)
|
33 |
- General Models Pre-Trained from Scratch (on 100B tokes):
|
|
|
84 |
}
|
85 |
```
|
86 |
|
87 |
+
[Adapt LLM to Domains](https://huggingface.co/papers/2309.09530)
|
88 |
```bibtex
|
89 |
@inproceedings{
|
90 |
cheng2024adapting,
|