Skip to content

Commit

Permalink
Merge branch 'FlagOpen:main' into chatglm3_6b
Browse files Browse the repository at this point in the history
  • Loading branch information
wangxichi authored Sep 5, 2024
2 parents 7290a9d + 3363e47 commit 71add5b
Show file tree
Hide file tree
Showing 1,190 changed files with 21,998 additions and 3,130 deletions.
29 changes: 24 additions & 5 deletions README.md
Original file line number Diff line number Diff line change
Expand Up @@ -79,7 +79,7 @@
<td>FP64算力</td>
<td>算力</td>
<td><a href="https://github.com/FlagOpen/FlagPerf/tree/main/base/benchmarks/computation-FP64/nvidia/A100">算子或原语</a>,<br><a href="https://github.com/FlagOpen/FlagPerf/tree/main/base/toolkits/computation-FP64/nvidia/A100">厂商专用工具</a></td>
<td><a href="https://github.com/FlagOpen/FlagPerf/tree/main/base/benchmarks/computation-FP64/metax">算子或原语</a></td>
<td>N/A</td>
<td>N/A</td>
</tr>
<tr>
Expand Down Expand Up @@ -144,15 +144,15 @@
<td>互联</td>
<td><a href="https://github.com/FlagOpen/FlagPerf/tree/main/base/benchmarks/interconnect-h2d/nvidia/A100">算子或原语</a>,<br><a href="https://github.com/FlagOpen/FlagPerf/tree/main/base/toolkits/interconnect-h2d/nvidia/A100">厂商专用工具</a></td>
<td>N/A</td>
<td>N/A</td>
<td><a href="https://github.com/FlagOpen/FlagPerf/tree/main/base/toolkits/interconnect-h2d/ascend">厂商专用工具</a></td>
</tr>
<tr>
<td>10</td>
<td>服务器内P2P直连</td>
<td>互联</td>
<td><a href="https://github.com/FlagOpen/FlagPerf/tree/main/base/benchmarks/interconnect-P2P_intraserver/nvidia/A100">算子或原语</a>,<br><a href="https://github.com/FlagOpen/FlagPerf/tree/main/base/toolkits/interconnect-P2P_intraserver/nvidia/A100">厂商专用工具</a></td>
<td>N/A</td>
<td>N/A</td>
<td><a href="https://github.com/FlagOpen/FlagPerf/tree/main/base/toolkits/interconnect-P2P_intraserver/ascend">厂商专用工具</a></td>
</tr>
<tr>
<td>11</td>
Expand Down Expand Up @@ -212,6 +212,12 @@
<td>nativetorch<br>flaggems</td>
<td><a href="https://github.com/FlagOpen/FlagPerf/tree/main/operation/benchmarks/linear/nvidia">A100_40_SXM</a></td>
</tr>
<tr>
<td>...</td>
<td>...</td>
<td>...</td>
<td>...</td>
</tr>
</tbody>
</table>

Expand Down Expand Up @@ -244,6 +250,7 @@
<td class="xl65" x:str>天数智芯</td>
<td class="xl65" x:str>腾讯九霄</td>
<td class="xl65" x:str>沐曦</td>
<td class="xl65" x:str>海飞科</td>
</tr>
<tr height="16.80" style='height:16.80pt;'>
<td class="xl65" x:str>1</td>
Expand All @@ -254,6 +261,7 @@
<td class="xl69" x:str>f16</td>
<td class="xl69" x:str>f16</td>
<td class="xl69" x:str>f32/f16</td>
<td class="xl69" x:str>N/A</td>
</tr>
<tr height="16.80" style='height:16.80pt;'>
<td class="xl65" x:str>2</td>
Expand All @@ -264,6 +272,7 @@
<td class="xl69" x:str>f16</td>
<td class="xl69" x:str>N/A</td>
<td class="xl69" x:str>f32/f16</td>
<td class="xl69" x:str>N/A</td>
</tr>
<tr height="16.80" style='height:16.80pt;'>
<td class="xl65" x:str>3</td>
Expand All @@ -274,6 +283,7 @@
<td class="xl69" x:str>N/A</td>
<td class="xl69" x:str>N/A</td>
<td class="xl69" x:str>f32/f16</td>
<td class="xl69" x:str>N/A</td>
</tr>
<tr height="16.80" style='height:16.80pt;'>
<td class="xl65" x:str>4</td>
Expand All @@ -284,6 +294,7 @@
<td class="xl69" x:str>f16</td>
<td class="xl69" x:str>N/A</td>
<td class="xl69" x:str>f32/f16</td>
<td class="xl69" x:str>N/A</td>
</tr>
<tr height="16.80" style='height:16.80pt;'>
<td class="xl65" x:str>5</td>
Expand All @@ -294,6 +305,7 @@
<td class="xl69" x:str>N/A</td>
<td class="xl69" x:str>N/A</td>
<td class="xl69" x:str>f32/f16</td>
<td class="xl69" x:str>N/A</td>
</tr>
<tr height="16.80" style='height:16.80pt;'>
<td class="xl65" x:str>6</td>
Expand All @@ -304,6 +316,7 @@
<td class="xl69" x:str>N/A</td>
<td class="xl69" x:str>N/A</td>
<td class="xl69" x:str>f32/f16</td>
<td class="xl69" x:str>N/A</td>
</tr>
<tr height="16.80" style='height:16.80pt;'>
<td class="xl65" x:str>7</td>
Expand All @@ -314,6 +327,7 @@
<td class="xl69" x:str>N/A</td>
<td class="xl69" x:str>N/A</td>
<td class="xl69" x:str>f32/f16</td>
<td class="xl69" x:str>f32/f16</td>
</tr>
<tr height="16.80" style='height:16.80pt;'>
<td class="xl65" x:str>8</td>
Expand All @@ -324,6 +338,7 @@
<td class="xl69" x:str>N/A</td>
<td class="xl69" x:str>N/A</td>
<td class="xl69" x:str>f16</td>
<td class="xl69" x:str>N/A</td>
</tr>
<tr height="16.80" style='height:16.80pt;'>
<td class="xl65" x:str>9</td>
Expand All @@ -334,6 +349,7 @@
<td class="xl69" x:str>N/A</td>
<td class="xl69" x:str>N/A</td>
<td class="xl69" x:str>f32/f16</td>
<td class="xl69" x:str>N/A</td>
</tr>
<tr height="16.80" style='height:16.80pt;'>
<td class="xl65" x:str>10</td>
Expand All @@ -344,7 +360,8 @@
<td class="xl69" x:str>N/A</td>
<td class="xl69" x:str>N/A</td>
<td class="xl69" x:str>N/A</td>
</tr><<tr height="16.80" style='height:16.80pt;'>
<td class="xl69" x:str>N/A</td>
</tr><tr height="16.80" style='height:16.80pt;'>
<td class="xl65" x:str>11</td>
<td class="xl65" height="33.60" style='height:33.60pt;border-right:none;border-bottom:none;' x:str><a href="https://github.com/FlagOpen/FlagPerf/tree/main/inference/benchmarks/llama3_8b_mmlu" style="text-decoration:none" target="_parent">LLaMA3-8B MMLU</td>
<td class="xl69" x:str>LLM</td>
Expand All @@ -353,7 +370,9 @@
<td class="xl69" x:str>N/A</td>
<td class="xl69" x:str>N/A</td>
<td class="xl69" x:str>N/A</td>
</tr></table>
<td class="xl69" x:str>N/A</td>
</tr>
</table>


## 如何使用FlagPerf进行AI硬件评测
Expand Down
Binary file modified assets/imgs/overview.png
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
44 changes: 44 additions & 0 deletions base/benchmarks/computation-BF16/cambricon/MLU/README.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,44 @@
# 参评AI芯片信息

* 厂商:Cambricon

## 服务器1

- 产品名称:/
- 产品型号:/
- TDP:/

# 所用服务器配置

* 服务器数量:1

## 服务器1

* 单服务器内使用卡数:8
* 服务器型号:/
* 操作系统版本:Ubuntu 22.04.1 LTS
* 操作系统内核:linux5.15.0-97-generic
* CPU:/
* docker版本:25.0.3
* 内存:2TiB
* 服务器间AI芯片直连规格及带宽:此评测样例无需服务器间通信

# 评测结果

## 核心评测结果

| 评测项 | BF16算力测试值(8卡平均) | BF16算力标定值(8卡平均) | 测试标定比例(8卡平均) |
| ---- | ---------------- | ---------------- | ------------- |
| 评测结果 | / | / | / |

## 能耗监控结果

| 监控项 | 系统平均功耗 | 系统最大功耗 | 系统功耗标准差 | 单机TDP | 单卡平均功耗(8卡平均) | 单卡最大功耗(8卡最大) | 单卡功耗标准差(8卡平均) | 单卡TDP |
| ---- | ------------ | ------------ | ------------- | ----- | ------------- | ------------- | -------------- | ----- |
| 监控结果 | / | / | / | / | / | / | / | / |

## 其他重要监控结果

| 监控项 | 系统平均CPU占用 | 系统平均内存占用 | 单卡平均温度(8卡平均) | 单卡平均显存占用(8卡平均) |
| ---- | --------------- | -------------- | ------------- | --------------- |
| 监控结果 | / | / | / |/ |
Original file line number Diff line number Diff line change
@@ -0,0 +1,6 @@
M: 10240
N: 10240
K: 10240
WARMUP: 100
ITERS: 50000
DIST_BACKEND: "cncl"
1 change: 1 addition & 0 deletions base/benchmarks/computation-BF16/cambricon/MLU/env.sh
Original file line number Diff line number Diff line change
@@ -0,0 +1 @@
echo "Cambricon PLACEHOLDER ENV.SH"
File renamed without changes.
7 changes: 7 additions & 0 deletions base/benchmarks/computation-BF16/main.py
Original file line number Diff line number Diff line change
Expand Up @@ -3,6 +3,13 @@
# Licensed under the Apache License, Version 2.0 (the "License")
#!/usr/bin/env python3
# -*- coding: UTF-8 -*-

# cambricon mlu import
try:
from torch_mlu.utils.model_transfer import transfer
except ImportError:
pass

import torch
import torch.distributed as dist
import os
Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -4,9 +4,9 @@

## 服务器1

- 产品名称:C500
- 产品型号:曦云®C500 64G
- TDP:350W
- 产品名称:C550
- 产品型号:曦云®C550 64G
- TDP:450W

# 所用服务器配置

Expand All @@ -15,10 +15,10 @@
## 服务器1

* 单服务器内使用卡数:8
* 服务器型号:同泰怡 G658V3
* 服务器型号:OAM C550-1500
* 操作系统版本:Ubuntu 20.04.6 LTS
* 操作系统内核:linux5.15.0-58-generic
* CPU:Montage Jintide(R) C8458P-176core
* CPU:Inter(R) Xeon(R) Plattinum 8480+
* docker版本:24.0.7
* 内存:2TiB
* 服务器间AI芯片直连规格及带宽:此评测样例无需服务器间通信
Expand All @@ -29,16 +29,16 @@

| 评测项 | BF16算力测试值(8卡平均) | BF16算力标定值(8卡平均) | 测试标定比例(8卡平均) |
| ---- | ---------------- | ---------------- | ------------- |
| 评测结果 | | | 94.2% |
| 评测结果 | | | 82.8% |

## 能耗监控结果

| 监控项 | 系统平均功耗 | 系统最大功耗 | 系统功耗标准差 | 单机TDP | 单卡平均功耗(8卡平均) | 单卡最大功耗(8卡最大) | 单卡功耗标准差(8卡平均) | 单卡TDP |
| ---- | ------------ | ------------ | ------------- | ----- | ------------- | ------------- | -------------- | ----- |
| 监控结果 | 1866.92W | 1998.0W | 88.62W | / | 62.5W | 68.0W | 5.5W | 350W |
| 监控结果 | 4207.5 | 4233.0 | 25.5 | / | 125.5W | 150.0W | 24.5W | 450W |

## 其他重要监控结果

| 监控项 | 系统平均CPU占用 | 系统平均内存占用 | 单卡平均温度(8卡平均) | 单卡平均显存占用(8卡平均) |
| ---- | --------------- | -------------- | ------------- | --------------- |
| 监控结果 | 3.097% | 1.313% | 36.0°C | 3.752% |
| 监控结果 | 0.784% | 0.55% | 35.5°C | 5.003% |
5 changes: 5 additions & 0 deletions base/benchmarks/computation-BF16/metax/C550/case_config.yaml
Original file line number Diff line number Diff line change
@@ -0,0 +1,5 @@
M: 6656
N: 2048
K: 4096
ITERS: 500
DIST_BACKEND: "nccl"
44 changes: 44 additions & 0 deletions base/benchmarks/computation-FP16/cambricon/MLU/README.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,44 @@
# 参评AI芯片信息

* 厂商:Cambricon

## 服务器1

- 产品名称:/
- 产品型号:/
- TDP:/

# 所用服务器配置

* 服务器数量:1

## 服务器1

* 单服务器内使用卡数:8
* 服务器型号:/
* 操作系统版本:Ubuntu 22.04.1 LTS
* 操作系统内核:linux5.15.0-97-generic
* CPU:/
* docker版本:25.0.3
* 内存:2TiB
* 服务器间AI芯片直连规格及带宽:此评测样例无需服务器间通信

# 评测结果

## 核心评测结果

| 评测项 | FP16算力测试值(8卡平均) | FP16算力标定值(8卡平均) | 测试标定比例(8卡平均) |
| ---- | ---------------- | ---------------- | ------------- |
| 评测结果 | / | / | / | |

## 能耗监控结果

| 监控项 | 系统平均功耗 | 系统最大功耗 | 系统功耗标准差 | 单机TDP | 单卡平均功耗(8卡平均) | 单卡最大功耗(8卡最大) | 单卡功耗标准差(8卡平均) | 单卡TDP |
| ---- | ------------ | ------------ | ------------- | ----- | ------------- | ------------- | -------------- | ----- |
| 监控结果 | / | / | / | / | / | / | / | / |

## 其他重要监控结果

| 监控项 | 系统平均CPU占用 | 系统平均内存占用 | 单卡平均温度(8卡平均) | 单卡平均显存占用(8卡平均) |
| ---- | --------------- | -------------- | ------------- | --------------- |
| 监控结果 | / | / | / |/ |
Original file line number Diff line number Diff line change
@@ -0,0 +1,4 @@
M: 10240
N: 10240
K: 10240
DIST_BACKEND: "cncl"
1 change: 1 addition & 0 deletions base/benchmarks/computation-FP16/cambricon/MLU/env.sh
Original file line number Diff line number Diff line change
@@ -0,0 +1 @@
echo "Cambricon PLACEHOLDER ENV.SH"
File renamed without changes.
7 changes: 7 additions & 0 deletions base/benchmarks/computation-FP16/main.py
Original file line number Diff line number Diff line change
Expand Up @@ -3,6 +3,13 @@
# Licensed under the Apache License, Version 2.0 (the "License")
#!/usr/bin/env python3
# -*- coding: UTF-8 -*-

# cambricon mlu import
try:
from torch_mlu.utils.model_transfer import transfer
except ImportError:
pass

import torch
import torch.distributed as dist
import os
Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -4,9 +4,9 @@

## 服务器1

- 产品名称:C500
- 产品型号:曦云®C500 64G
- TDP:350W
- 产品名称:C550
- 产品型号:曦云®C550 64G
- TDP:450W

# 所用服务器配置

Expand All @@ -15,10 +15,10 @@
## 服务器1

* 单服务器内使用卡数:8
* 服务器型号:同泰怡 G658V3
* 服务器型号:OAM C550-1500
* 操作系统版本:Ubuntu 20.04.6 LTS
* 操作系统内核:linux5.15.0-58-generic
* CPU:Montage Jintide(R) C8458P-176core
* CPU:Inter(R) Xeon(R) Plattinum 8480+
* docker版本:24.0.7
* 内存:2TiB
* 服务器间AI芯片直连规格及带宽:此评测样例无需服务器间通信
Expand All @@ -27,18 +27,18 @@

## 核心评测结果

| 评测项 | BF16算力测试值(8卡平均) | BF16算力标定值(8卡平均) | 测试标定比例(8卡平均) |
| 评测项 | FP16算力测试值(8卡平均) | FP16算力标定值(8卡平均) | 测试标定比例(8卡平均) |
| ---- | ---------------- | ---------------- | ------------- |
| 评测结果 | | | 88.55% |
| 评测结果 | | | 83.5% |

## 能耗监控结果

| 监控项 | 系统平均功耗 | 系统最大功耗 | 系统功耗标准差 | 单机TDP | 单卡平均功耗(8卡平均) | 单卡最大功耗(8卡最大) | 单卡功耗标准差(8卡平均) | 单卡TDP |
| ---- | ------------ | ------------ | ------------- | ----- | ------------- | ------------- | -------------- | ----- |
| 监控结果 | 2012.0W | 3486.0W | 450.54W | / | 69.0W | 81.0W | 12.0W | 350W |
| 监控结果 | 4182.0W | 4182.0W | 0.0W | / | 112.5W | 124.0W | 11.5W | 450W |

## 其他重要监控结果

| 监控项 | 系统平均CPU占用 | 系统平均内存占用 | 单卡平均温度(8卡平均) | 单卡平均显存占用(8卡平均) |
| ---- | --------------- | -------------- | ------------- | --------------- |
| 监控结果 | 3.423% | 1.34% | 36.5°C | 5.148% |
| 监控结果 | 0.872% | 0.55% | 34.5°C | 4.71% |
5 changes: 5 additions & 0 deletions base/benchmarks/computation-FP16/metax/C550/case_config.yaml
Original file line number Diff line number Diff line change
@@ -0,0 +1,5 @@
M: 6656
N: 2048
K: 4096
ITERS: 500
DIST_BACKEND: "nccl"
Loading

0 comments on commit 71add5b

Please sign in to comment.