更新qwen2.5-72B-32K性能yaml,启用dw掩盖提升性能

This commit is contained in:
z30032740
2025-01-02 15:50:23 +08:00
parent 464962f99a
commit c1b52ba7fa
3 changed files with 22 additions and 2 deletions

View File

@@ -150,11 +150,12 @@ context:
device_id: 0
ascend_config:
precision_mode: "must_keep_origin_dtype"
parallel_speed_up_json_path: "./research/qwen2_5/parallel_speed_up_72B_32K.json" # Replace with a real path when needed
# model config
model:
model_config:
use_attn_mask_compression: True
use_attn_mask_compression: False
type: LlamaConfig
batch_size: 1
seq_length: 32768

View File

@@ -0,0 +1,3 @@
{
"matmul_grad_comm_overlap": true
}

View File

@@ -227,7 +227,23 @@ train_dataset: &train_dataset
dynamic_batch: True
```
4. 启动微调:
4. 配置并行加速
若模型的yaml中有类似如下的配置
```yaml
context:
ascend_config:
parallel_speed_up_json_path: "/path/to/parallel_speed_up.json" # Replace with a real path when needed
```
这是在使用`parallel_speed_up`文件(须是`json`格式)去配置一些并行加速特性,以获得一些性能上的提升。实际使用时,请把`json`文件的路径修改为实际值。
> `parallel_speed_up`文件中各配置项的含义详见[parallel_speed_up说明](https://www.mindspore.cn/docs/zh-CN/master/api_python/mindspore/mindspore.set_context.html#:~:text=parallel_speed_up_json_path)。
例如,`finetune_qwen2.5_72B_32K.yaml`中使用了`parallel_speed_up_72B_32K.json`,其中配置了`"matmul_grad_comm_overlap": true`。
5. 启动微调:
```shell
cd research/qwen2_5