update GPU benchmark

vtuber-plan · Mar 2, 2024 · 6440063 · 6440063
1 parent c6caefe
commit 6440063
Showing 1 changed file with 8 additions and 4 deletions.
diff --git a/README.md b/README.md
@@ -4,6 +4,7 @@ Easy Language Model Trainer
 
 ## Let's benchmark GPUs LLM training with katheryne
 
+### Training settings for 7B Llama (100 steps)
 ```
 Stage: Pretrain
 Model: meta-llama/Llama-2-7b-hf
@@ -17,7 +18,10 @@ dtype: bf16
 lora: {"r": 16, "target_modules": ["q_proj", "v_proj"]}
 ```
 
-|     GPU     |  Time |     Memory   |
-|-------------|-------|--------------|
-|  RTX 3090   |       |   14,928MiB  |
-|  Tesla A800 |       |              |
+|              GPU             |  Time   |     Memory   |
+|------------------------------|---------|--------------|
+|    NVIDIA A800 80GB PCIe     |  00:41  |   14,850MiB  |
+|   NVIDIA GeForce RTX 4090    |  01:02  |   15,078MiB  |
+|      Iluvatar BI-V150        |  01:09  |   22,798MiB  |
+|   NVIDIA GeForce RTX 3090    |  01:36  |   14,928MiB  |
+