diff --git a/PyTorch/built-in/rl/VeRL_for_PyTorch/README.md b/PyTorch/built-in/rl/VeRL_for_PyTorch/README.md index 0ed8d871a401fa4b44e0240e07072b0412ad023d..2ca07fb51b018da39f41fd97f0192a24b83cbb5b 100644 --- a/PyTorch/built-in/rl/VeRL_for_PyTorch/README.md +++ b/PyTorch/built-in/rl/VeRL_for_PyTorch/README.md @@ -261,16 +261,18 @@ verl‌是一个集SFT(监督学习)与RL(强化学习)于一体的灵 **表 2** 训练结果展示表 -| MODEL | NAME | throughput | MAX Training TimeSteps | -|:-----------------------|:------------------------|:----------:|:----------------------:| -| Qwen2.5-VL-3B-Instruct | 8p-竞品A | 739.453 | 60 | -| Qwen2.5-VL-3B-Instruct | 8P Atlas 200T A2 Box16 | 349.013 | 60 | -| Qwen2.5-VL-7B-Instruct | 8p-竞品A | 568.452 | 60 | -| Qwen2.5-VL-7B-Instruct | 16P Atlas 200T A2 Box16 | 216.796 | 60 | -| Qwen2.5-7B-Instruct | 8p-竞品A | 323.872 | 35 | -| Qwen2.5-7B-Instruct | 16P Atlas 200T A2 Box16 | 190.617 | 35 | -| Qwen2.5-32B-Instruct | 16p-竞品A | 79.022 | 105 | -| Qwen2.5-32B-Instruct | 32P Atlas 200T A2 Box16 | 54.162 | 105 | +| MODEL | NAME | throughput | MAX Training TimeSteps | +|:------------------------|:------------------------|:----------:|:----------------------:| +| Qwen2.5-VL-3B-Instruct | 8p-竞品A | 739.453 | 60 | +| Qwen2.5-VL-3B-Instruct | 8P Atlas 200T A2 Box16 | 349.013 | 60 | +| Qwen2.5-VL-7B-Instruct | 8p-竞品A | 568.452 | 60 | +| Qwen2.5-VL-7B-Instruct | 16P Atlas 200T A2 Box16 | 216.796 | 60 | +| Qwen2.5-VL-32B-Instruct | 16p-竞品A | 109.497 | 60 | +| Qwen2.5-VL-32B-Instruct | 32P Atlas 200T A2 Box16 | 62.2283 | 60 | +| Qwen2.5-7B-Instruct | 8p-竞品A | 323.872 | 35 | +| Qwen2.5-7B-Instruct | 16P Atlas 200T A2 Box16 | 190.617 | 35 | +| Qwen2.5-32B-Instruct | 16p-竞品A | 79.022 | 105 | +| Qwen2.5-32B-Instruct | 32P Atlas 200T A2 Box16 | 54.162 | 105 | # 公网地址说明