From 19e8d03440e09f91d973fb712d7408f4a753e261 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?=E5=8F=B6=E4=B8=80=E5=B8=86?= Date: Mon, 28 Apr 2025 15:54:32 +0800 Subject: [PATCH 1/6] =?UTF-8?q?=E6=9B=B4=E6=96=B0qwen2.5=E7=9A=84readme?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- MindIE/LLM/Qwen2.5/Qwen2.5-14B-Instruct/README.md | 11 ++++++----- MindIE/LLM/Qwen2.5/Qwen2.5-7B-Instruct/README.md | 11 ++++++----- 2 files changed, 12 insertions(+), 10 deletions(-) diff --git a/MindIE/LLM/Qwen2.5/Qwen2.5-14B-Instruct/README.md b/MindIE/LLM/Qwen2.5/Qwen2.5-14B-Instruct/README.md index da9b8fc076..98458baadb 100644 --- a/MindIE/LLM/Qwen2.5/Qwen2.5-14B-Instruct/README.md +++ b/MindIE/LLM/Qwen2.5/Qwen2.5-14B-Instruct/README.md @@ -2,7 +2,7 @@ - 千问（Qwen2.5）大语言模型能够理解和生成文本，应用于智能客服、内容生成、问答系统等多个场景，助力企业智能化升级。 -- 此代码仓中实现了一套基于NPU硬件的qwen2.5推理模型。配合加速库使用，旨在NPU上获得极致的推理性能。 +- 此代码仓中实现了一套基于NPU硬件的Qwen2.5推理模型。配合加速库使用，旨在NPU上获得极致的推理性能。 # 特性矩阵 @@ -10,7 +10,7 @@ | 模型及参数量 | 800I A2 Tensor Parallelism | 300I DUO Tensor Parallelism | FP16 | BF16 | Flash Attention | Paged Attention | W8A8量化 | W8A16量化 | KV cache量化 | 稀疏量化 | MOE量化 | MindIE Service | TGI | 长序列 | prefix_cache | FA3量化 | functioncall | Multi LoRA| | ----------------- |----------------------------|-----------------------------| ---- | ---- | --------------- | --------------- | -------- | --------- | ------------ | -------- | ------- | -------------- | --- | ------ | ---------- | --- | --- | --- | -| Qwen2.5-14B | 支持world size ,2,4,8 | 支持world size 1,2,4,8 | √ | √(800I A2/32G/64G) | × | √ | √(800I A2/32G/64G) | × | × | √(300I DUO) | × | √ | × | √ | √ | × | √ | x | +| Qwen2.5-14B | 支持world size ,2,4,8 | 支持world size 1,2,4,8 | √ | √(800I A2/32G/64G) | × | √ | √(800I A2/32G/64G) | × | × | √(300I DUO) | × | √ | × | × | √ | × | √ | x | 注：表中所示支持的world size为对话测试可跑通的配置，实际运行时还需考虑输入序列长度带来的显存占用。 @@ -22,7 +22,7 @@ | --------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------- | | working_dir | 加速库及模型库下载后放置的目录 | | llm_path | 模型仓所在路径。若使用编译好的包，则路径为`/usr/local/Ascend/atb-models`；若使用 gitee 下载的代码，则路径为`${working_dir}/MindIE-LLM/examples/atb_models` | -| script_path | 脚本所在路径；qwen2.5 的工作脚本所在路径为`${llm_path}/examples/models/qwen` | +| script_path | 脚本所在路径；Qwen2.5 的工作脚本所在路径为`${llm_path}/examples/models/qwen` | | weight_path | 模型权重路径 | | rank_table_path | Rank table文件路径 | @@ -139,7 +139,7 @@ source /usr/local/Ascend/mindie/set_env.sh #### 参数说明（需要到脚本中修改） -根据硬件设备不同请参考下表修改run_pa.sh再运行 +根据硬件设备不同请参考下表修改`/usr/local/Ascend/atb-models/examples/models/qwen/run_pa.sh`再运行 | 参数名称 | 含义 | 800I A2推荐值 | 300I DUO推荐值 | | ------------------------- | ----------------------------------------- | ---------------- | ---------------- | @@ -227,6 +227,7 @@ pkill -9 -f 'mindie|python' ``` 3. 执行命令： ``` +export MINDIE_LOG_TO_STDOUT=1 bash run.sh pa_[data_type] performance [case_pair] [batch_size] ([prefill_batch_size]) [model_name] ([is_chat_model]) [weight_dir] [world_size] ``` 参数说明： @@ -414,7 +415,7 @@ source /usr/local/Ascend/ascend-toolkit/set_env.sh source /usr/local/Ascend/nnal/atb/set_env.sh source /usr/local/Ascend/atb-models/set_env.sh source /usr/local/Ascend/mindie/set_env.sh - +export MINDIE_LOG_TO_STDOUT=1 chmod 640 /usr/local/lib/python3.11/site-packages/mindiebenchmark/config/config.json ``` diff --git a/MindIE/LLM/Qwen2.5/Qwen2.5-7B-Instruct/README.md b/MindIE/LLM/Qwen2.5/Qwen2.5-7B-Instruct/README.md index bd56c4a350..3877818565 100644 --- a/MindIE/LLM/Qwen2.5/Qwen2.5-7B-Instruct/README.md +++ b/MindIE/LLM/Qwen2.5/Qwen2.5-7B-Instruct/README.md @@ -2,7 +2,7 @@ - 千问（Qwen2.5）大语言模型能够理解和生成文本，应用于智能客服、内容生成、问答系统等多个场景，助力企业智能化升级。 -- 此代码仓中实现了一套基于NPU硬件的qwen2.5推理模型。配合加速库使用，旨在NPU上获得极致的推理性能。 +- 此代码仓中实现了一套基于NPU硬件的Qwen2.5推理模型。配合加速库使用，旨在NPU上获得极致的推理性能。 # 特性矩阵 @@ -10,7 +10,7 @@ | 模型及参数量 | 800I A2 Tensor Parallelism | 300I DUO Tensor Parallelism | FP16 | BF16 | Flash Attention | Paged Attention | W8A8量化 | W8A16量化 | KV cache量化 | 稀疏量化 | MOE量化 | MindIE Service | TGI | 长序列 | prefix_cache | FA3量化 | functioncall | Multi LoRA| | ----------------- |----------------------------|-----------------------------| ---- | ---- | --------------- | --------------- | -------- | --------- | ------------ | -------- | ------- | -------------- | --- | ------ | ---------- | --- | --- | --- | -| Qwen2.5-7B | 支持world size 1,2,4,8 | 支持world size 1,2,4,8 | √ | √(800I A2/32G/64G) | × | √ | √(800I A2/32G/64G) | × | × | √(300I DUO) | × | √ | × | √ | √ | × | √ | x | +| Qwen2.5-7B | 支持world size 1,2,4,8 | 支持world size 1,2,4,8 | √ | √(800I A2/32G/64G) | × | √ | √(800I A2/32G/64G) | × | × | √(300I DUO) | × | √ | × | × | √ | × | √ | x | 注：表中所示支持的world size为对话测试可跑通的配置，实际运行时还需考虑输入序列长度带来的显存占用。 @@ -22,7 +22,7 @@ | --------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------- | | working_dir | 加速库及模型库下载后放置的目录 | | llm_path | 模型仓所在路径。若使用编译好的包，则路径为`/usr/local/Ascend/atb-models`；若使用 gitee 下载的代码，则路径为`${working_dir}/MindIE-LLM/examples/atb_models` | -| script_path | 脚本所在路径；qwen2.5 的工作脚本所在路径为`${llm_path}/examples/models/qwen` | +| script_path | 脚本所在路径；Qwen2.5 的工作脚本所在路径为`${llm_path}/examples/models/qwen` | | weight_path | 模型权重路径 | | rank_table_path | Rank table文件路径 | @@ -139,7 +139,7 @@ source /usr/local/Ascend/mindie/set_env.sh #### 参数说明（需要到脚本中修改） -根据硬件设备不同请参考下表修改run_pa.sh再运行 +根据硬件设备不同请参考下表修改`/usr/local/Ascend/atb-models/examples/models/qwen/run_pa.sh`再运行 | 参数名称 | 含义 | 800I A2推荐值 | 300I DUO推荐值 | | ------------------------- | ----------------------------------------- | ---------------- | ---------------- | @@ -227,6 +227,7 @@ pkill -9 -f 'mindie|python' ``` 3. 执行命令： ``` +export MINDIE_LOG_TO_STDOUT=1 bash run.sh pa_[data_type] performance [case_pair] [batch_size] ([prefill_batch_size]) [model_name] ([is_chat_model]) [weight_dir] [world_size] ``` 参数说明： @@ -414,7 +415,7 @@ source /usr/local/Ascend/ascend-toolkit/set_env.sh source /usr/local/Ascend/nnal/atb/set_env.sh source /usr/local/Ascend/atb-models/set_env.sh source /usr/local/Ascend/mindie/set_env.sh - +export MINDIE_LOG_TO_STDOUT=1 chmod 640 /usr/local/lib/python3.11/site-packages/mindiebenchmark/config/config.json ``` -- Gitee From 1cedbc0eed6f23a896e3e2c505a2581c92e59db9 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?=E5=8F=B6=E4=B8=80=E5=B8=86?= Date: Mon, 28 Apr 2025 17:29:38 +0800 Subject: [PATCH 2/6] =?UTF-8?q?=E8=B0=83=E6=95=B4=E9=A1=BA=E5=BA=8F?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- .../Qwen2.5/Qwen2.5-14B-Instruct/README.md | 88 +++++++++---------- .../LLM/Qwen2.5/Qwen2.5-7B-Instruct/README.md | 88 +++++++++---------- 2 files changed, 88 insertions(+), 88 deletions(-) diff --git a/MindIE/LLM/Qwen2.5/Qwen2.5-14B-Instruct/README.md b/MindIE/LLM/Qwen2.5/Qwen2.5-14B-Instruct/README.md index 98458baadb..4526d24b76 100644 --- a/MindIE/LLM/Qwen2.5/Qwen2.5-14B-Instruct/README.md +++ b/MindIE/LLM/Qwen2.5/Qwen2.5-14B-Instruct/README.md @@ -26,6 +26,50 @@ | weight_path | 模型权重路径 | | rank_table_path | Rank table文件路径 | +## 加载镜像 +前往[昇腾社区/开发资源](https://www.hiascend.com/developer/ascendhub/detail/af85b724a7e5469ebd7ea13c3439d48f)下载`1.0.0-300I-Duo-py311-openeuler24.03-lts`或`1.0.0-800I-A2-py311-openeuler24.03-lts`镜像，下载镜像前需要申请权限，耐心等待权限申请通过后，根据指南下载对应镜像文件。 + +完成之后，请使用docker images命令确认查找具体镜像名称与标签。 + +## 启动容器 +- 执行以下命令启动容器（参考）： +```sh +docker run -itd --privileged --name= {容器名称} --net=host \ + --shm-size 500g \ + --device=/dev/davinci0 \ + --device=/dev/davinci1 \ + --device=/dev/davinci2 \ + --device=/dev/davinci3 \ + --device=/dev/davinci4 \ + --device=/dev/davinci5 \ + --device=/dev/davinci6 \ + --device=/dev/davinci7 \ + --device=/dev/davinci_manager \ + --device=/dev/hisi_hdc \ + --device /dev/devmm_svm \ + -v /usr/local/Ascend/driver:/usr/local/Ascend/driver \ + -v /usr/local/Ascend/firmware:/usr/local/Ascend/firmware \ + -v /usr/local/sbin/npu-smi:/usr/local/sbin/npu-smi \ + -v /usr/local/sbin:/usr/local/sbin \ + -v /etc/hccn.conf:/etc/hccn.conf \ + -v {/权重路径:/权重路径} \ + -v {swr.cn-south-1.myhuaweicloud.com/ascendhub/mindie:1.0.0-XXX-800I-A2-arm64-py3.11（根据加载的镜像名称修改）} \ + bash +``` +#### 进入容器 +- 执行以下命令进入容器（参考）： +```sh +docker exec -it {容器名称} bash +``` + +#### 设置基础环境变量 +``` +source /usr/local/Ascend/ascend-toolkit/set_env.sh +source /usr/local/Ascend/nnal/atb/set_env.sh +source /usr/local/Ascend/atb-models/set_env.sh +source /usr/local/Ascend/mindie/set_env.sh +``` + ## 模型权重 **权重下载** @@ -90,50 +134,6 @@ torchrun --nproc_per_node 2 -m examples.convert.model_slim.sparse_compressor --model_path /data1/weights/model_slim/Qwen-14b_w8a8s --save_directory /data1/weights/model_slim/Qwen-14b_w8a8sc ``` -## 加载镜像 -前往[昇腾社区/开发资源](https://www.hiascend.com/developer/ascendhub/detail/af85b724a7e5469ebd7ea13c3439d48f)下载`1.0.0-300I-Duo-py311-openeuler24.03-lts`或`1.0.0-800I-A2-py311-openeuler24.03-lts`镜像，下载镜像前需要申请权限，耐心等待权限申请通过后，根据指南下载对应镜像文件。 - -完成之后，请使用docker images命令确认查找具体镜像名称与标签。 - -## 启动容器 -- 执行以下命令启动容器（参考）： -```sh -docker run -itd --privileged --name= {容器名称} --net=host \ - --shm-size 500g \ - --device=/dev/davinci0 \ - --device=/dev/davinci1 \ - --device=/dev/davinci2 \ - --device=/dev/davinci3 \ - --device=/dev/davinci4 \ - --device=/dev/davinci5 \ - --device=/dev/davinci6 \ - --device=/dev/davinci7 \ - --device=/dev/davinci_manager \ - --device=/dev/hisi_hdc \ - --device /dev/devmm_svm \ - -v /usr/local/Ascend/driver:/usr/local/Ascend/driver \ - -v /usr/local/Ascend/firmware:/usr/local/Ascend/firmware \ - -v /usr/local/sbin/npu-smi:/usr/local/sbin/npu-smi \ - -v /usr/local/sbin:/usr/local/sbin \ - -v /etc/hccn.conf:/etc/hccn.conf \ - -v {/权重路径:/权重路径} \ - -v {swr.cn-south-1.myhuaweicloud.com/ascendhub/mindie:1.0.0-XXX-800I-A2-arm64-py3.11（根据加载的镜像名称修改）} \ - bash -``` -#### 进入容器 -- 执行以下命令进入容器（参考）： -```sh -docker exec -it {容器名称} bash -``` - -#### 设置基础环境变量 -``` -source /usr/local/Ascend/ascend-toolkit/set_env.sh -source /usr/local/Ascend/nnal/atb/set_env.sh -source /usr/local/Ascend/atb-models/set_env.sh -source /usr/local/Ascend/mindie/set_env.sh -``` - ## 纯模型推理： 【使用场景】使用相同输入长度和相同输出长度，构造多Batch去测试纯模型性能 diff --git a/MindIE/LLM/Qwen2.5/Qwen2.5-7B-Instruct/README.md b/MindIE/LLM/Qwen2.5/Qwen2.5-7B-Instruct/README.md index 3877818565..6c8a79b9b4 100644 --- a/MindIE/LLM/Qwen2.5/Qwen2.5-7B-Instruct/README.md +++ b/MindIE/LLM/Qwen2.5/Qwen2.5-7B-Instruct/README.md @@ -26,6 +26,50 @@ | weight_path | 模型权重路径 | | rank_table_path | Rank table文件路径 | +## 加载镜像 +前往[昇腾社区/开发资源](https://www.hiascend.com/developer/ascendhub/detail/af85b724a7e5469ebd7ea13c3439d48f)下载`1.0.0-300I-Duo-py311-openeuler24.03-lts`或`1.0.0-800I-A2-py311-openeuler24.03-lts`镜像，下载镜像前需要申请权限，耐心等待权限申请通过后，根据指南下载对应镜像文件。 + +完成之后，请使用docker images命令确认查找具体镜像名称与标签。 + +## 启动容器 +- 执行以下命令启动容器（参考）： +```sh +docker run -itd --privileged --name= {容器名称} --net=host \ + --shm-size 500g \ + --device=/dev/davinci0 \ + --device=/dev/davinci1 \ + --device=/dev/davinci2 \ + --device=/dev/davinci3 \ + --device=/dev/davinci4 \ + --device=/dev/davinci5 \ + --device=/dev/davinci6 \ + --device=/dev/davinci7 \ + --device=/dev/davinci_manager \ + --device=/dev/hisi_hdc \ + --device /dev/devmm_svm \ + -v /usr/local/Ascend/driver:/usr/local/Ascend/driver \ + -v /usr/local/Ascend/firmware:/usr/local/Ascend/firmware \ + -v /usr/local/sbin/npu-smi:/usr/local/sbin/npu-smi \ + -v /usr/local/sbin:/usr/local/sbin \ + -v /etc/hccn.conf:/etc/hccn.conf \ + -v {/权重路径:/权重路径} \ + -v {swr.cn-south-1.myhuaweicloud.com/ascendhub/mindie:1.0.0-XXX-800I-A2-arm64-py3.11（根据加载的镜像名称修改）} \ + bash +``` +#### 进入容器 +- 执行以下命令进入容器（参考）： +```sh +docker exec -it {容器名称} bash +``` + +#### 设置基础环境变量 +``` +source /usr/local/Ascend/ascend-toolkit/set_env.sh +source /usr/local/Ascend/nnal/atb/set_env.sh +source /usr/local/Ascend/atb-models/set_env.sh +source /usr/local/Ascend/mindie/set_env.sh +``` + ## 模型权重 **权重下载** @@ -90,50 +134,6 @@ torchrun --nproc_per_node 2 -m examples.convert.model_slim.sparse_compressor --model_path /data1/weights/model_slim/Qwen-7b_w8a8s --save_directory /data1/weights/model_slim/Qwen-7b_w8a8sc ``` -## 加载镜像 -前往[昇腾社区/开发资源](https://www.hiascend.com/developer/ascendhub/detail/af85b724a7e5469ebd7ea13c3439d48f)下载`1.0.0-300I-Duo-py311-openeuler24.03-lts`或`1.0.0-800I-A2-py311-openeuler24.03-lts`镜像，下载镜像前需要申请权限，耐心等待权限申请通过后，根据指南下载对应镜像文件。 - -完成之后，请使用docker images命令确认查找具体镜像名称与标签。 - -## 启动容器 -- 执行以下命令启动容器（参考）： -```sh -docker run -itd --privileged --name= {容器名称} --net=host \ - --shm-size 500g \ - --device=/dev/davinci0 \ - --device=/dev/davinci1 \ - --device=/dev/davinci2 \ - --device=/dev/davinci3 \ - --device=/dev/davinci4 \ - --device=/dev/davinci5 \ - --device=/dev/davinci6 \ - --device=/dev/davinci7 \ - --device=/dev/davinci_manager \ - --device=/dev/hisi_hdc \ - --device /dev/devmm_svm \ - -v /usr/local/Ascend/driver:/usr/local/Ascend/driver \ - -v /usr/local/Ascend/firmware:/usr/local/Ascend/firmware \ - -v /usr/local/sbin/npu-smi:/usr/local/sbin/npu-smi \ - -v /usr/local/sbin:/usr/local/sbin \ - -v /etc/hccn.conf:/etc/hccn.conf \ - -v {/权重路径:/权重路径} \ - -v {swr.cn-south-1.myhuaweicloud.com/ascendhub/mindie:1.0.0-XXX-800I-A2-arm64-py3.11（根据加载的镜像名称修改）} \ - bash -``` -#### 进入容器 -- 执行以下命令进入容器（参考）： -```sh -docker exec -it {容器名称} bash -``` - -#### 设置基础环境变量 -``` -source /usr/local/Ascend/ascend-toolkit/set_env.sh -source /usr/local/Ascend/nnal/atb/set_env.sh -source /usr/local/Ascend/atb-models/set_env.sh -source /usr/local/Ascend/mindie/set_env.sh -``` - ## 纯模型推理： 【使用场景】使用相同输入长度和相同输出长度，构造多Batch去测试纯模型性能 -- Gitee From 6e5363bd65ec9571681a62fbe3df9e22ab6bec6c Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?=E5=8F=B6=E4=B8=80=E5=B8=86?= Date: Mon, 28 Apr 2025 17:47:24 +0800 Subject: [PATCH 3/6] add --- .../Qwen2.5/Qwen2.5-14B-Instruct/README.md | 49 ++++++++++++++++++- .../LLM/Qwen2.5/Qwen2.5-7B-Instruct/README.md | 48 +++++++++++++++++- 2 files changed, 94 insertions(+), 3 deletions(-) diff --git a/MindIE/LLM/Qwen2.5/Qwen2.5-14B-Instruct/README.md b/MindIE/LLM/Qwen2.5/Qwen2.5-14B-Instruct/README.md index 4526d24b76..bdad9c7cec 100644 --- a/MindIE/LLM/Qwen2.5/Qwen2.5-14B-Instruct/README.md +++ b/MindIE/LLM/Qwen2.5/Qwen2.5-14B-Instruct/README.md @@ -197,7 +197,7 @@ bash run.sh pa_[data_type] [dataset] ([shots]) [batch_size] [model_name] ([is_ch ``` 参数说明： 1. `data_type`：为数据类型，根据权重目录下config.json的data_type选择bf16或者fp16，例如：pa_bf16。 -2. `dataset`：可选full_BoolQ、full_CEval等，相关数据集可至[魔乐社区MindIE](https://modelers.cn/MindIE/data.git)下载，（下载之前，需要申请加入组织，下载之后拷贝到/usr/local/Ascend/atb-models/tests/modeltest/路径下）CEval与MMLU等数据集需要设置`shots`（通常设为5）。 +2. `dataset`：可选full_BoolQ、full_CEval等，相关数据集需要自行下载，（参考[附录]，下载之后拷贝到/usr/local/Ascend/atb-models/tests/modeltest/路径下）CEval与MMLU等数据集需要设置`shots`（通常设为5）。 3. `batch_size`：为`batch数`。 4. `model_name`：为`qwen`。 5. `is_chat_model`：为`是否支持对话模式，若传入此参数，则进入对话模式`。 @@ -444,7 +444,52 @@ benchmark \ ``` ModelName，ModelPath需要与mindie-service里的config.json里的一致。样例仅供参考，请根据实际情况调整参数。 - +## 附录 + +#### 数据集下载 + +- 首先，需要在test/modeltest路径下新建名为temp_data的文件目录，然后在temp_data文件目录下新建对应数据集文件目录: + +| 支持数据集 | 目录名称 | +|---------------|---------------| +| BoolQ | boolq | +| HumanEval | humaneval | +| HumanEval_X | humaneval_x | +| GSM8K | gsm8k | +| LongBench | longbench | +| MMLU | mmlu | +| NeedleBench | needlebench | +| VideoBench | VideoBench | +| Vocalsound | Vocalsound | +| TextVQA | TextVQA | +| TruthfulQA | truthfulqa | + +- 获取数据集：需要访问huggingface和github的对应网址，手动下载对应数据集 + +| 支持数据集 | 下载地址 | +|----------------|-----------------------------| +| BoolQ |[dev.jsonl](https://storage.cloud.google.com/boolq/dev.jsonl)| +| HumanEval |[humaneval](https://github.com/openai/human-eval/raw/refs/heads/master/data/HumanEval.jsonl.gz)| +|HumanEval_X|[cpp](https://huggingface.co/datasets/THUDM/humaneval-x/tree/main/data/cpp/data)
[java](https://huggingface.co/datasets/THUDM/humaneval-x/tree/main/data/java/data)
[go](https://huggingface.co/datasets/THUDM/humaneval-x/tree/main/data/go/data)
[js](https://huggingface.co/datasets/THUDM/humaneval-x/tree/main/data/js/data)
[python](https://huggingface.co/datasets/THUDM/humaneval-x/tree/main/data/python/data)| +| GSM8K |[gsm8k](https://github.com/openai/grade-school-math/blob/master/grade_school_math/data/test.jsonl)| +| LongBench |[longbench](https://huggingface.co/datasets/THUDM/LongBench/resolve/main/data.zip)| +| MMLU |[mmlu](https://people.eecs.berkeley.edu/~hendrycks/data.tar)| +|NeedleBench|[PaulGrahamEssays](https://huggingface.co/datasets/opencompass/NeedleBench/tree/main)
[multi_needle_reasoning_en](https://huggingface.co/datasets/opencompass/NeedleBench/tree/main)
[multi_needle_reasoning_zh](https://huggingface.co/datasets/opencompass/NeedleBench/tree/main)
[names](https://huggingface.co/datasets/opencompass/NeedleBench/tree/main)
[needles](https://huggingface.co/datasets/opencompass/NeedleBench/tree/main)
[zh_finance](https://huggingface.co/datasets/opencompass/NeedleBench/tree/main)
[zh_game](https://huggingface.co/datasets/opencompass/NeedleBench/tree/main)
[zh_general](https://huggingface.co/datasets/opencompass/NeedleBench/tree/main)
[zh_government](https://huggingface.co/datasets/opencompass/NeedleBench/tree/main)
[zh_movie](https://huggingface.co/datasets/opencompass/NeedleBench/tree/main)
[zh_tech](https://huggingface.co/datasets/opencompass/NeedleBench/tree/main)| +|TextVQA|[train_val_images.zip](https://dl.fbaipublicfiles.com/textvqa/images/train_val_images.zip)
[textvqa_val.jsonl](https://ofasys-wlcb.oss-cn-wulanchabu.aliyuncs.com/Qwen-VL/evaluation/textvqa/textvqa_val.jsonl)
[textvqa_val_annotations.json](https://ofasys-wlcb.oss-cn-wulanchabu.aliyuncs.com/Qwen-VL/evaluation/textvqa/textvqa_val_annotations.json)
| +|VideoBench|[Eval_QA/](https://github.com/PKU-YuanGroup/Video-Bench)
[Video-Bench](https://huggingface.co/datasets/LanguageBind/Video-Bench/tree/main)
| +|VocalSound|[VocalSound 16kHz Version](https://www.dropbox.com/s/c5ace70qh1vbyzb/vs_release_16k.zip?dl=1)
| +|TruthfulQA|[truthfulqa](https://huggingface.co/datasets/domenicrosati/TruthfulQA/tree/main)| + +- 将对应下载的数据集文件放置在对应的数据集目录下，并在modeltest根目录`MindIE-LLM/examples/atb_models/tests/modeltest`下执行： + +```bash +python3 scripts/data_prepare.py [可选参数] +``` + +| 参数名 | 含义 | +|--------|------------------------------| +| dataset_name | 可选，需要下载的数据集名称，支持的数据集列表参见[**功能**]章节，多个名称以','隔开 | +| remove_cache | 可选，是否在下载前清除数据集缓存 | ## FAQ diff --git a/MindIE/LLM/Qwen2.5/Qwen2.5-7B-Instruct/README.md b/MindIE/LLM/Qwen2.5/Qwen2.5-7B-Instruct/README.md index 6c8a79b9b4..3f490b2c07 100644 --- a/MindIE/LLM/Qwen2.5/Qwen2.5-7B-Instruct/README.md +++ b/MindIE/LLM/Qwen2.5/Qwen2.5-7B-Instruct/README.md @@ -197,7 +197,7 @@ bash run.sh pa_[data_type] [dataset] ([shots]) [batch_size] [model_name] ([is_ch ``` 参数说明： 1. `data_type`：为数据类型，根据权重目录下config.json的data_type选择bf16或者fp16，例如：pa_bf16。 -2. `dataset`：可选full_BoolQ、full_CEval等，相关数据集可至[魔乐社区MindIE](https://modelers.cn/MindIE/data.git)下载，（下载之前，需要申请加入组织，下载之后拷贝到/usr/local/Ascend/atb-models/tests/modeltest/路径下）CEval与MMLU等数据集需要设置`shots`（通常设为5）。 +2. `dataset`：可选full_BoolQ、full_CEval等，相关数据集需要自行下载，（参考[附录]，下载之后拷贝到/usr/local/Ascend/atb-models/tests/modeltest/路径下）CEval与MMLU等数据集需要设置`shots`（通常设为5）。 3. `batch_size`：为`batch数`。 4. `model_name`：为`qwen`。 5. `is_chat_model`：为`是否支持对话模式，若传入此参数，则进入对话模式`。 @@ -444,6 +444,52 @@ benchmark \ ``` ModelName，ModelPath需要与mindie-service里的config.json里的一致。样例仅供参考，请根据实际情况调整参数。 +## 附录 + +#### 数据集下载 + +- 首先，需要在test/modeltest路径下新建名为temp_data的文件目录，然后在temp_data文件目录下新建对应数据集文件目录: + +| 支持数据集 | 目录名称 | +|---------------|---------------| +| BoolQ | boolq | +| HumanEval | humaneval | +| HumanEval_X | humaneval_x | +| GSM8K | gsm8k | +| LongBench | longbench | +| MMLU | mmlu | +| NeedleBench | needlebench | +| VideoBench | VideoBench | +| Vocalsound | Vocalsound | +| TextVQA | TextVQA | +| TruthfulQA | truthfulqa | + +- 获取数据集：需要访问huggingface和github的对应网址，手动下载对应数据集 + +| 支持数据集 | 下载地址 | +|----------------|-----------------------------| +| BoolQ |[dev.jsonl](https://storage.cloud.google.com/boolq/dev.jsonl)| +| HumanEval |[humaneval](https://github.com/openai/human-eval/raw/refs/heads/master/data/HumanEval.jsonl.gz)| +|HumanEval_X|[cpp](https://huggingface.co/datasets/THUDM/humaneval-x/tree/main/data/cpp/data)
[java](https://huggingface.co/datasets/THUDM/humaneval-x/tree/main/data/java/data)
[go](https://huggingface.co/datasets/THUDM/humaneval-x/tree/main/data/go/data)
[js](https://huggingface.co/datasets/THUDM/humaneval-x/tree/main/data/js/data)
[python](https://huggingface.co/datasets/THUDM/humaneval-x/tree/main/data/python/data)| +| GSM8K |[gsm8k](https://github.com/openai/grade-school-math/blob/master/grade_school_math/data/test.jsonl)| +| LongBench |[longbench](https://huggingface.co/datasets/THUDM/LongBench/resolve/main/data.zip)| +| MMLU |[mmlu](https://people.eecs.berkeley.edu/~hendrycks/data.tar)| +|NeedleBench|[PaulGrahamEssays](https://huggingface.co/datasets/opencompass/NeedleBench/tree/main)
[multi_needle_reasoning_en](https://huggingface.co/datasets/opencompass/NeedleBench/tree/main)
[multi_needle_reasoning_zh](https://huggingface.co/datasets/opencompass/NeedleBench/tree/main)
[names](https://huggingface.co/datasets/opencompass/NeedleBench/tree/main)
[needles](https://huggingface.co/datasets/opencompass/NeedleBench/tree/main)
[zh_finance](https://huggingface.co/datasets/opencompass/NeedleBench/tree/main)
[zh_game](https://huggingface.co/datasets/opencompass/NeedleBench/tree/main)
[zh_general](https://huggingface.co/datasets/opencompass/NeedleBench/tree/main)
[zh_government](https://huggingface.co/datasets/opencompass/NeedleBench/tree/main)
[zh_movie](https://huggingface.co/datasets/opencompass/NeedleBench/tree/main)
[zh_tech](https://huggingface.co/datasets/opencompass/NeedleBench/tree/main)| +|TextVQA|[train_val_images.zip](https://dl.fbaipublicfiles.com/textvqa/images/train_val_images.zip)
[textvqa_val.jsonl](https://ofasys-wlcb.oss-cn-wulanchabu.aliyuncs.com/Qwen-VL/evaluation/textvqa/textvqa_val.jsonl)
[textvqa_val_annotations.json](https://ofasys-wlcb.oss-cn-wulanchabu.aliyuncs.com/Qwen-VL/evaluation/textvqa/textvqa_val_annotations.json)
| +|VideoBench|[Eval_QA/](https://github.com/PKU-YuanGroup/Video-Bench)
[Video-Bench](https://huggingface.co/datasets/LanguageBind/Video-Bench/tree/main)
| +|VocalSound|[VocalSound 16kHz Version](https://www.dropbox.com/s/c5ace70qh1vbyzb/vs_release_16k.zip?dl=1)
| +|TruthfulQA|[truthfulqa](https://huggingface.co/datasets/domenicrosati/TruthfulQA/tree/main)| + +- 将对应下载的数据集文件放置在对应的数据集目录下，并在modeltest根目录`MindIE-LLM/examples/atb_models/tests/modeltest`下执行： + +```bash +python3 scripts/data_prepare.py [可选参数] +``` + +| 参数名 | 含义 | +|--------|------------------------------| +| dataset_name | 可选，需要下载的数据集名称，支持的数据集列表参见[**功能**]章节，多个名称以','隔开 | +| remove_cache | 可选，是否在下载前清除数据集缓存 | ## FAQ -- Gitee From db4bc891c6cf1e62b5071cdb25ef7a2c42fdba43 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?=E5=8F=B6=E4=B8=80=E5=B8=86?= Date: Mon, 28 Apr 2025 17:48:33 +0800 Subject: [PATCH 4/6] add --- MindIE/LLM/Qwen2.5/Qwen2.5-7B-Instruct/README.md | 1 - 1 file changed, 1 deletion(-) diff --git a/MindIE/LLM/Qwen2.5/Qwen2.5-7B-Instruct/README.md b/MindIE/LLM/Qwen2.5/Qwen2.5-7B-Instruct/README.md index 3f490b2c07..591db1b6b7 100644 --- a/MindIE/LLM/Qwen2.5/Qwen2.5-7B-Instruct/README.md +++ b/MindIE/LLM/Qwen2.5/Qwen2.5-7B-Instruct/README.md @@ -491,7 +491,6 @@ python3 scripts/data_prepare.py [可选参数] | dataset_name | 可选，需要下载的数据集名称，支持的数据集列表参见[**功能**]章节，多个名称以','隔开 | | remove_cache | 可选，是否在下载前清除数据集缓存 | - ## FAQ - 对话测试实际执行的 Python 文件为`${llm_path}/examples/run_pa.py` -- Gitee From 0d928dbb5fcebb9fab5576a515f7decf26fad1ac Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?=E5=8F=B6=E4=B8=80=E5=B8=86?= Date: Fri, 16 May 2025 12:27:32 +0800 Subject: [PATCH 5/6] fix --- MindIE/LLM/Qwen2.5/Qwen2.5-14B-Instruct/README.md | 8 ++++++-- MindIE/LLM/Qwen2.5/Qwen2.5-7B-Instruct/README.md | 6 +++++- 2 files changed, 11 insertions(+), 3 deletions(-) diff --git a/MindIE/LLM/Qwen2.5/Qwen2.5-14B-Instruct/README.md b/MindIE/LLM/Qwen2.5/Qwen2.5-14B-Instruct/README.md index bdad9c7cec..1d59ffc840 100644 --- a/MindIE/LLM/Qwen2.5/Qwen2.5-14B-Instruct/README.md +++ b/MindIE/LLM/Qwen2.5/Qwen2.5-14B-Instruct/README.md @@ -10,7 +10,7 @@ | 模型及参数量 | 800I A2 Tensor Parallelism | 300I DUO Tensor Parallelism | FP16 | BF16 | Flash Attention | Paged Attention | W8A8量化 | W8A16量化 | KV cache量化 | 稀疏量化 | MOE量化 | MindIE Service | TGI | 长序列 | prefix_cache | FA3量化 | functioncall | Multi LoRA| | ----------------- |----------------------------|-----------------------------| ---- | ---- | --------------- | --------------- | -------- | --------- | ------------ | -------- | ------- | -------------- | --- | ------ | ---------- | --- | --- | --- | -| Qwen2.5-14B | 支持world size ,2,4,8 | 支持world size 1,2,4,8 | √ | √(800I A2/32G/64G) | × | √ | √(800I A2/32G/64G) | × | × | √(300I DUO) | × | √ | × | × | √ | × | √ | x | +| Qwen2.5-14B | 支持world size 2,4,8 | 支持world size 1,2,4,8 | √ | √(800I A2/32G/64G) | × | √ | √(800I A2/32G/64G) | × | × | √(300I DUO) | × | √ | × | × | √ | × | √ | x | 注：表中所示支持的world size为对话测试可跑通的配置，实际运行时还需考虑输入序列长度带来的显存占用。 @@ -453,6 +453,8 @@ ModelName，ModelPath需要与mindie-service里的config.json里的一致。样 | 支持数据集 | 目录名称 | |---------------|---------------| | BoolQ | boolq | +| CEval | ceval | +| CMMLU | cmmlu | | HumanEval | humaneval | | HumanEval_X | humaneval_x | | GSM8K | gsm8k | @@ -469,6 +471,8 @@ ModelName，ModelPath需要与mindie-service里的config.json里的一致。样 | 支持数据集 | 下载地址 | |----------------|-----------------------------| | BoolQ |[dev.jsonl](https://storage.cloud.google.com/boolq/dev.jsonl)| +| CEval |[ceval-exam](https://huggingface.co/datasets/ceval/ceval-exam/resolve/main/ceval-exam.zip)| +| CMMLU |[cmmlu](https://huggingface.co/datasets/haonan-li/cmmlu/resolve/main/cmmlu_v1_0_1.zip)| | HumanEval |[humaneval](https://github.com/openai/human-eval/raw/refs/heads/master/data/HumanEval.jsonl.gz)| |HumanEval_X|[cpp](https://huggingface.co/datasets/THUDM/humaneval-x/tree/main/data/cpp/data)
[java](https://huggingface.co/datasets/THUDM/humaneval-x/tree/main/data/java/data)
[go](https://huggingface.co/datasets/THUDM/humaneval-x/tree/main/data/go/data)
[js](https://huggingface.co/datasets/THUDM/humaneval-x/tree/main/data/js/data)
[python](https://huggingface.co/datasets/THUDM/humaneval-x/tree/main/data/python/data)| | GSM8K |[gsm8k](https://github.com/openai/grade-school-math/blob/master/grade_school_math/data/test.jsonl)| @@ -488,7 +492,7 @@ python3 scripts/data_prepare.py [可选参数] | 参数名 | 含义 | |--------|------------------------------| -| dataset_name | 可选，需要下载的数据集名称，支持的数据集列表参见[**功能**]章节，多个名称以','隔开 | +| dataset_name | 可选，需要下载的数据集名称，多个名称以','隔开 | | remove_cache | 可选，是否在下载前清除数据集缓存 | ## FAQ diff --git a/MindIE/LLM/Qwen2.5/Qwen2.5-7B-Instruct/README.md b/MindIE/LLM/Qwen2.5/Qwen2.5-7B-Instruct/README.md index 591db1b6b7..9c26eb1699 100644 --- a/MindIE/LLM/Qwen2.5/Qwen2.5-7B-Instruct/README.md +++ b/MindIE/LLM/Qwen2.5/Qwen2.5-7B-Instruct/README.md @@ -453,6 +453,8 @@ ModelName，ModelPath需要与mindie-service里的config.json里的一致。样 | 支持数据集 | 目录名称 | |---------------|---------------| | BoolQ | boolq | +| CEval | ceval | +| CMMLU | cmmlu | | HumanEval | humaneval | | HumanEval_X | humaneval_x | | GSM8K | gsm8k | @@ -469,6 +471,8 @@ ModelName，ModelPath需要与mindie-service里的config.json里的一致。样 | 支持数据集 | 下载地址 | |----------------|-----------------------------| | BoolQ |[dev.jsonl](https://storage.cloud.google.com/boolq/dev.jsonl)| +| CEval |[ceval-exam](https://huggingface.co/datasets/ceval/ceval-exam/resolve/main/ceval-exam.zip)| +| CMMLU |[cmmlu](https://huggingface.co/datasets/haonan-li/cmmlu/resolve/main/cmmlu_v1_0_1.zip)| | HumanEval |[humaneval](https://github.com/openai/human-eval/raw/refs/heads/master/data/HumanEval.jsonl.gz)| |HumanEval_X|[cpp](https://huggingface.co/datasets/THUDM/humaneval-x/tree/main/data/cpp/data)
[java](https://huggingface.co/datasets/THUDM/humaneval-x/tree/main/data/java/data)
[go](https://huggingface.co/datasets/THUDM/humaneval-x/tree/main/data/go/data)
[js](https://huggingface.co/datasets/THUDM/humaneval-x/tree/main/data/js/data)
[python](https://huggingface.co/datasets/THUDM/humaneval-x/tree/main/data/python/data)| | GSM8K |[gsm8k](https://github.com/openai/grade-school-math/blob/master/grade_school_math/data/test.jsonl)| @@ -488,7 +492,7 @@ python3 scripts/data_prepare.py [可选参数] | 参数名 | 含义 | |--------|------------------------------| -| dataset_name | 可选，需要下载的数据集名称，支持的数据集列表参见[**功能**]章节，多个名称以','隔开 | +| dataset_name | 可选，需要下载的数据集名称，多个名称以','隔开 | | remove_cache | 可选，是否在下载前清除数据集缓存 | ## FAQ -- Gitee From e1bd76462de0c0796a882e6a99f00c26e866df3a Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?=E5=8F=B6=E4=B8=80=E5=B8=86?= Date: Thu, 22 May 2025 16:09:04 +0800 Subject: [PATCH 6/6] =?UTF-8?q?=E6=9B=B4=E6=AD=A3=E5=90=AF=E5=8A=A8?= =?UTF-8?q?=E5=AE=B9=E5=99=A8=E8=84=9A=E6=9C=AC?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- MindIE/LLM/Qwen2.5/Qwen2.5-14B-Instruct/README.md | 2 +- MindIE/LLM/Qwen2.5/Qwen2.5-7B-Instruct/README.md | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-) diff --git a/MindIE/LLM/Qwen2.5/Qwen2.5-14B-Instruct/README.md b/MindIE/LLM/Qwen2.5/Qwen2.5-14B-Instruct/README.md index 1d59ffc840..e7a5dbfcc2 100644 --- a/MindIE/LLM/Qwen2.5/Qwen2.5-14B-Instruct/README.md +++ b/MindIE/LLM/Qwen2.5/Qwen2.5-14B-Instruct/README.md @@ -53,7 +53,7 @@ docker run -itd --privileged --name= {容器名称} --net=host \ -v /usr/local/sbin:/usr/local/sbin \ -v /etc/hccn.conf:/etc/hccn.conf \ -v {/权重路径:/权重路径} \ - -v {swr.cn-south-1.myhuaweicloud.com/ascendhub/mindie:1.0.0-XXX-800I-A2-arm64-py3.11（根据加载的镜像名称修改）} \ + {swr.cn-south-1.myhuaweicloud.com/ascendhub/mindie:1.0.0-XXX-800I-A2-arm64-py3.11（根据加载的镜像名称修改）} \ bash ``` #### 进入容器 diff --git a/MindIE/LLM/Qwen2.5/Qwen2.5-7B-Instruct/README.md b/MindIE/LLM/Qwen2.5/Qwen2.5-7B-Instruct/README.md index 9c26eb1699..2913ea5faa 100644 --- a/MindIE/LLM/Qwen2.5/Qwen2.5-7B-Instruct/README.md +++ b/MindIE/LLM/Qwen2.5/Qwen2.5-7B-Instruct/README.md @@ -53,7 +53,7 @@ docker run -itd --privileged --name= {容器名称} --net=host \ -v /usr/local/sbin:/usr/local/sbin \ -v /etc/hccn.conf:/etc/hccn.conf \ -v {/权重路径:/权重路径} \ - -v {swr.cn-south-1.myhuaweicloud.com/ascendhub/mindie:1.0.0-XXX-800I-A2-arm64-py3.11（根据加载的镜像名称修改）} \ + {swr.cn-south-1.myhuaweicloud.com/ascendhub/mindie:1.0.0-XXX-800I-A2-arm64-py3.11（根据加载的镜像名称修改）} \ bash ``` #### 进入容器 -- Gitee