Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,28 @@
|
|
| 1 |
---
|
| 2 |
license: apache-2.0
|
| 3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
license: apache-2.0
|
| 3 |
---
|
| 4 |
+
|
| 5 |
+
# etri-xainlp/llama2-13b-lima-sft-dpo
|
| 6 |
+
|
| 7 |
+
## Model Details
|
| 8 |
+
|
| 9 |
+
**Model Developers** ETRI xainlp team
|
| 10 |
+
|
| 11 |
+
**Input** text only.
|
| 12 |
+
|
| 13 |
+
**Output** text only.
|
| 14 |
+
|
| 15 |
+
**Model Architecture**
|
| 16 |
+
|
| 17 |
+
**Base Model** [meta-llama/Llama-13b-hf](https://huggingface.co/meta-llama/Llama-2-13b-hf)
|
| 18 |
+
|
| 19 |
+
**Training Dataset**
|
| 20 |
+
|
| 21 |
+
- fully sft: 650k instruction-following set
|
| 22 |
+
|
| 23 |
+
- lima sft: 280k instruction-following set
|
| 24 |
+
|
| 25 |
+
- dpo+lora: 90k user preference set
|
| 26 |
+
|
| 27 |
+
- We use A100 GPU 80GB * 7, when training.
|
| 28 |
+
|