Update README.md
Browse files
README.md
CHANGED
|
@@ -59,6 +59,9 @@ SFT LoRA обучение было выполнено на **двух NVIDIA A10
|
|
| 59 |
- Liger Kernel (swiglu, fused linear xentropy)
|
| 60 |
|
| 61 |
**GPU hours**: ~384 of NVIDIA A100
|
|
|
|
|
|
|
|
|
|
| 62 |
|
| 63 |
### Training configuration / Конфигурация обучения
|
| 64 |
**The model was trained using MyLLM framework:**
|
|
|
|
| 59 |
- Liger Kernel (swiglu, fused linear xentropy)
|
| 60 |
|
| 61 |
**GPU hours**: ~384 of NVIDIA A100
|
| 62 |
+
**GPU mem**:
|
| 63 |
+
- Stage 1: 50-55GB of VRAM (both GPUs)
|
| 64 |
+
- Stage 2: 79GB of VRAM (both GPUs)
|
| 65 |
|
| 66 |
### Training configuration / Конфигурация обучения
|
| 67 |
**The model was trained using MyLLM framework:**
|