JingweiZuo
commited on
Commit
•
02688fc
1
Parent(s):
124f971
Update README.md
Browse files
README.md
CHANGED
@@ -154,8 +154,8 @@ print(tokenizer.decode(outputs[0]))
|
|
154 |
## Training Data
|
155 |
|
156 |
Falcon-Mamba has been trained with ~ 5,500 GT mainly coming from [Refined-Web](https://huggingface.co/datasets/tiiuae/falcon-refinedweb), a large volume web-only dataset filtered and deduplicated.
|
157 |
-
Similar to the others [Falcon](https://huggingface.co/tiiuae/falcon-11B) suite models, Falcon-Mamba has been trained leveraging a multi-stage training strategy to increase the context-length
|
158 |
-
Moreover, inspired by the Curriculum Learning
|
159 |
Note that at inference the context-length is not relevant as the Mamba architecture has no limit on long range dependency.
|
160 |
At the last training stage, small portion of high-quality curated data was used to further enhance performance.
|
161 |
|
|
|
154 |
## Training Data
|
155 |
|
156 |
Falcon-Mamba has been trained with ~ 5,500 GT mainly coming from [Refined-Web](https://huggingface.co/datasets/tiiuae/falcon-refinedweb), a large volume web-only dataset filtered and deduplicated.
|
157 |
+
Similar to the others [Falcon](https://huggingface.co/tiiuae/falcon-11B) suite models, Falcon-Mamba has been trained leveraging a multi-stage training strategy to increase the context-length from 2,048 to 8,192.
|
158 |
+
Moreover, inspired by the concept of Curriculum Learning, we carefully selected data mixtures throughout the training stages, considering both data diversity and complexity.
|
159 |
Note that at inference the context-length is not relevant as the Mamba architecture has no limit on long range dependency.
|
160 |
At the last training stage, small portion of high-quality curated data was used to further enhance performance.
|
161 |
|