Update README.md
Browse files
README.md
CHANGED
|
@@ -1,21 +1,22 @@
|
|
| 1 |
---
|
| 2 |
license: cc-by-nc-4.0
|
| 3 |
---
|
| 4 |
-
KoLLaVA : Korean Large Language and Vision Assistant (feat. LLaVA)
|
| 5 |
This model is a large multimodal model (LMM) that combines the LLM(LLaMA-2-7b-ko) with visual encoder of CLIP(ViT-14), trained on Korean visual-instruction dataset using QLoRA.
|
| 6 |
|
| 7 |
Detail codes are available at [KoLLaVA](https://github.com/tabtoyou/KoLLaVA/tree/main) github repository
|
| 8 |
|
| 9 |
-
Training hyperparameters
|
| 10 |
-
learning rate : 2e-4
|
| 11 |
-
train_batch_size: 16
|
| 12 |
-
distributed_type: multi-GPU (RTX3090 24G)
|
| 13 |
-
num_devices: 4
|
| 14 |
-
gradient_accumulation_steps: 2
|
| 15 |
-
total_train_batch_size: 128
|
| 16 |
-
total_eval_batch_size: 4
|
| 17 |
-
lr_scheduler_type: cosine
|
| 18 |
-
num_epochs: 1
|
| 19 |
-
lora_enable: True
|
| 20 |
-
bits: 4
|
|
|
|
| 21 |
Model License: cc-by-nc-4.0
|
|
|
|
| 1 |
---
|
| 2 |
license: cc-by-nc-4.0
|
| 3 |
---
|
| 4 |
+
## KoLLaVA : Korean Large Language and Vision Assistant (feat. LLaVA)
|
| 5 |
This model is a large multimodal model (LMM) that combines the LLM(LLaMA-2-7b-ko) with visual encoder of CLIP(ViT-14), trained on Korean visual-instruction dataset using QLoRA.
|
| 6 |
|
| 7 |
Detail codes are available at [KoLLaVA](https://github.com/tabtoyou/KoLLaVA/tree/main) github repository
|
| 8 |
|
| 9 |
+
- Training hyperparameters
|
| 10 |
+
- learning rate : 2e-4
|
| 11 |
+
- train_batch_size: 16
|
| 12 |
+
- distributed_type: multi-GPU (RTX3090 24G)
|
| 13 |
+
- num_devices: 4
|
| 14 |
+
- gradient_accumulation_steps: 2
|
| 15 |
+
- total_train_batch_size: 128
|
| 16 |
+
- total_eval_batch_size: 4
|
| 17 |
+
- lr_scheduler_type: cosine
|
| 18 |
+
- num_epochs: 1
|
| 19 |
+
- lora_enable: True
|
| 20 |
+
- bits: 4
|
| 21 |
+
|
| 22 |
Model License: cc-by-nc-4.0
|