diff --git a/doc/en/train_performance.md b/doc/en/train_performance.md index 635bd0e..ff1a846 100644 --- a/doc/en/train_performance.md +++ b/doc/en/train_performance.md @@ -98,6 +98,7 @@ Test configuration: | :-------: | :-----: | | branch | develop | | commit id | 2b984ff | +| tag | v0.2.1dev20231121 | | GPU | A800 | | Checkpoint| True | | micro_bsz | 1 | @@ -128,7 +129,7 @@ parallel = dict( ) ``` -Pre-training test: +Pre-training & Fine-tuning test: |model|Number of GPU|zero1|tp|pp|fsdp|GPU Memory (GB)| |:-:|:-:|:-:|:-:|:-:|:-:|:-:| | 7B | 3 | -1 | 1 | 3 |False| 75 | @@ -154,4 +155,4 @@ Web_demo test: |model|GPU|GPU Memory (GB)|System Memory (MB)| |:-:|:-:|:-:|:-:| | 7B | A800 | 14.5 | 2465 | -| 7B | A800 | 39 | 9547 | +| 20B | A800 | 39 | 9547 | diff --git a/doc/train_performance.md b/doc/train_performance.md index ef174f9..65ed237 100644 --- a/doc/train_performance.md +++ b/doc/train_performance.md @@ -94,6 +94,7 @@ InternLM中`zero1`的配置决定了优化器状态的分配范围。 | :-------: | :-----: | | 分支 | develop | | commit id | 2b984ff | +| tag | v0.2.1dev20231121 | | 显卡 | A800 | | 重计算 | True | | micro_bsz | 1 | @@ -124,7 +125,7 @@ parallel = dict( ) ``` -预训练测试: +预训练和微调测试: |模型|卡数|zero1|tp|pp|fsdp|显存(GB)| |:-:|:-:|:-:|:-:|:-:|:-:|:-:| | 7B | 3 | -1 | 1 | 3 |False| 75 | @@ -150,4 +151,4 @@ Web_demo 测试: |模型|显卡|显存(GB)|内存(MB)| |:-:|:-:|:-:|:-:| | 7B | A800 | 14.5 | 2465 | -| 7B | A800 | 39 | 9547 | \ No newline at end of file +| 20B | A800 | 39 | 9547 | \ No newline at end of file