issue/896 - 为c++和python中的tensor添加打印函数 #930
Open
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
pengcheng888
commented
Jan 15, 2026
|
|
||
| import infinicore.context as context | ||
| import infinicore.nn as nn | ||
| from infinicore._tensor_str import printoptions, set_printoptions |
Collaborator
Author
There was a problem hiding this comment.
torch中把print的一些 函数调用放到了_tensor_str.py文件中，于是我也新建了一个同样的文件 _tensor_str.py
pengcheng888
commented
Jan 15, 2026
| "strided_from_blob", | ||
| "zeros", | ||
| "set_printoptions", | ||
| "printoptions", |
Collaborator
Author
There was a problem hiding this comment.
torch中存在一下两个函数
"set_printoptions"：全局配置的函数
"printoptions"： 临时配置的函数
pengcheng888
commented
Jan 15, 2026
| std::string repr = m1.extra_repr(); | ||
| spdlog::debug("Linear module representation: {}", repr); | ||
|
|
||
| // Test forward with residual connection |
Collaborator
Author
There was a problem hiding this comment.
c++的linear有个版移除了一个函数，导致c++测试编译不过。于是，在src/infinicore-test/test_nn_module.cc中注释或删除了传递的参数
pengcheng888
commented
Jan 15, 2026
|
|
||
| // Prepare test data (stored outside loop to ensure lifetime) | ||
| bool bool_data[4] = {true, false, true, false}; | ||
| int8_t i8_data[4] = {-128, -64, 32, 127}; |
Collaborator
Author
There was a problem hiding this comment.
由于infinicore.Tensor没有rand函数，没法生成随机数，所以就事前固定了测试数据
pengcheng888
commented
Jan 15, 2026
| set_installdir(os.getenv("INFINI_ROOT") or (os.getenv(is_host("windows") and "HOMEPATH" or "HOME") .. "/.infini")) | ||
| target_end() | ||
|
|
||
| target("infinicore-test") |
Collaborator
Author
There was a problem hiding this comment.
c++的infinicore的测试脚本出错，于是添加上了缺少的"infinicore_cpp_api"依赖
pengcheng888
commented
Jan 15, 2026
| template <typename T> | ||
| T item_impl(const std::byte *data, DataType dtype) { | ||
| switch (dtype) { | ||
| case DataType::F16: { |
Collaborator
Author
There was a problem hiding this comment.
由于c++没有原生的F16和BF16格式数据，为了能cout打印，把F16和BF16转成了float类型去显示
pengcheng888
force-pushed
the
issue/896
branch
2 times, most recently
from
665708e to
cb10145
Compare
pengcheng888
force-pushed
the
issue/896
branch
from
cb10145 to
c63bfa2
Compare
PanZezhong1725
approved these changes
Jan 21, 2026
pengcheng888
commented
Jan 21, 2026
|
|
||
| namespace infinicore { | ||
|
|
||
| std::string toString(const DataType &dtype) { |
Collaborator
Author
There was a problem hiding this comment.
为了数据类型的显示与torch对齐，修改了原来的名字
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
目标版本
main
功能描述
打印的核心逻辑参考了 https://github.com/xtensor-stack/xtensor， 核心代码在io.cc文件中
c++和python中的tensor添加打印函数，可以全局或临时配置一下参数：
(1) 触发省略的阈值：数据量过多时，显示部分数据；
(2) 每行显示字符数量：显示宽度合适，自动换行；
(3) 精度：设置显示的小数点位数；
(4) edgeitems: 省略显示时，设置显示的数据数量
c++的cout函数支持的类型： BOOL，I8，I16，I32，I64，U8，U16，U32，U64，BF16，F16，F32，F64
python的print函数支持的类型： BOOL，I8，I16，I32，I64，U8，BF16，F16，F32，F64
已测试平台：
nvidia moore metax iluvatar hygon
python代码std::cout测试：
临时修改后，自动恢复原始配置
全局修改后，配置生效
print不同的数据类型
c++代码print测试：
临时修改后，自动恢复原始配置
全局修改后，配置生效
cout不同的数据类型
在infinilm中使用：
