yuxiang630
commited on
Commit
•
cc7bf2c
1
Parent(s):
b120b50
refactor: use hyperlinks for images
Browse files
README.md
CHANGED
@@ -86,19 +86,17 @@ model-index:
|
|
86 |
|
87 |
# StarCoder2-Instruct: Self-Aligned, Transparent, and Fully Permissive
|
88 |
|
89 |
-
|
90 |
-
<img src="https://huggingface.co/datasets/bigcode/admin_private/resolve/main/starcoder2_banner.png" alt="SC2" width="900" height="600">
|
91 |
-
</center> -->
|
92 |
|
93 |
## Model Summary
|
94 |
|
95 |
We introduce StarCoder2-15B-Instruct-v0.1, the very first entirely self-aligned code Large Language Model (LLM) trained with a fully permissive and transparent pipeline. Our open-source pipeline uses StarCoder2-15B to generate thousands of instruction-response pairs, which are then used to fine-tune StarCoder-15B itself without any human annotations or distilled data from huge and proprietary LLMs.
|
96 |
|
97 |
-
- **Model:** [bigcode/
|
98 |
- **Code:** [bigcode-project/starcoder2-self-align](https://github.com/bigcode-project/starcoder2-self-align)
|
99 |
- **Dataset:** [bigcode/self-oss-instruct-sc2-exec-filter-50k](https://huggingface.co/datasets/bigcode/self-oss-instruct-sc2-exec-filter-50k/)
|
100 |
|
101 |
-
![self-alignment pipeline](
|
102 |
|
103 |
## Use
|
104 |
|
@@ -177,14 +175,15 @@ The model also inherits the bias, risks, and limitations from its base StarCoder
|
|
177 |
|
178 |
## Evaluation on EvalPlus, LiveCodeBench, and DS-1000
|
179 |
|
180 |
-
![EvalPlus](
|
181 |
|
182 |
-
![LiveCodeBench and DS-1000](
|
183 |
|
184 |
## Training Details
|
185 |
|
186 |
### Hyperparameters
|
187 |
|
|
|
188 |
- **Learning rate:** 1e-5
|
189 |
- **Epoch:** 4
|
190 |
- **Batch size:** 64
|
|
|
86 |
|
87 |
# StarCoder2-Instruct: Self-Aligned, Transparent, and Fully Permissive
|
88 |
|
89 |
+
![Banner](https://huggingface.co/datasets/bigcode/starcoder2-instruct-assets/resolve/main/banner.png)
|
|
|
|
|
90 |
|
91 |
## Model Summary
|
92 |
|
93 |
We introduce StarCoder2-15B-Instruct-v0.1, the very first entirely self-aligned code Large Language Model (LLM) trained with a fully permissive and transparent pipeline. Our open-source pipeline uses StarCoder2-15B to generate thousands of instruction-response pairs, which are then used to fine-tune StarCoder-15B itself without any human annotations or distilled data from huge and proprietary LLMs.
|
94 |
|
95 |
+
- **Model:** [bigcode/starcoder2-15b-instruct-v0.1](https://huggingface.co/bigcode/starcoder2-instruct-15b-v0.1)
|
96 |
- **Code:** [bigcode-project/starcoder2-self-align](https://github.com/bigcode-project/starcoder2-self-align)
|
97 |
- **Dataset:** [bigcode/self-oss-instruct-sc2-exec-filter-50k](https://huggingface.co/datasets/bigcode/self-oss-instruct-sc2-exec-filter-50k/)
|
98 |
|
99 |
+
![self-alignment pipeline](https://huggingface.co/datasets/bigcode/starcoder2-instruct-assets/resolve/main/method.png)
|
100 |
|
101 |
## Use
|
102 |
|
|
|
175 |
|
176 |
## Evaluation on EvalPlus, LiveCodeBench, and DS-1000
|
177 |
|
178 |
+
![EvalPlus](https://huggingface.co/datasets/bigcode/starcoder2-instruct-assets/resolve/main/evalplus.png)
|
179 |
|
180 |
+
![LiveCodeBench and DS-1000](https://huggingface.co/datasets/bigcode/starcoder2-instruct-assets/resolve/main/lcb-ds1000.png)
|
181 |
|
182 |
## Training Details
|
183 |
|
184 |
### Hyperparameters
|
185 |
|
186 |
+
- **Optimizer:** Adafactor
|
187 |
- **Learning rate:** 1e-5
|
188 |
- **Epoch:** 4
|
189 |
- **Batch size:** 64
|