Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[WeeklyReport] ooooo-create 2024.08.26~2024.10.29 #398

Open
wants to merge 4 commits into
base: main
Choose a base branch
from
Open
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Original file line number Diff line number Diff line change
@@ -0,0 +1,30 @@
### 姓名

方国勇

### 实习项目

PIR 专项

### 本周工作

- PIR在分布式下的单测修复工作

给分布式下单测加入 `deprecated` 目录,删除部分无用单测

- https://github.com/PaddlePaddle/Paddle/pull/67747
- https://github.com/PaddlePaddle/Paddle/pull/67798
- https://github.com/PaddlePaddle/Paddle/pull/67805
- https://github.com/PaddlePaddle/Paddle/pull/67854
- https://github.com/PaddlePaddle/Paddle/pull/67857
- https://github.com/PaddlePaddle/Paddle/pull/67862
- https://github.com/PaddlePaddle/Paddle/pull/67902
- https://github.com/PaddlePaddle/Paddle/pull/67995
- https://github.com/PaddlePaddle/Paddle/pull/68012
- https://github.com/PaddlePaddle/Paddle/pull/68013

### 下周工作

- 继续分布式下的单测修复工作

### 导师点评
Original file line number Diff line number Diff line change
@@ -0,0 +1,13 @@
### 姓名

方国勇

### 实习项目

PIR 专项

### 本周工作

### 下周工作

### 导师点评
Original file line number Diff line number Diff line change
@@ -0,0 +1,21 @@
### 姓名

方国勇

### 实习项目

PIR 专项

### 本周工作

- 学习 [CINN 动态 shape 符号推导](https://github.com/PaddlePaddle/Paddle/issues/66444)
- 本地完成 Linux 下 CINN 的编译
- 尝试完成 `det`, `is_empty`, `eye` 的符号推导接口
- `pd_op.h`, `pd_op.cc` 是在 cmake 阶段自动生成的,`pd_op.h` 都会自动生成*声明代码*,.cc 文件根据 ops.yaml 进行生成*实现代码*,所以需要重新 cmake
- 发现 ci 上单测通过,但是新增代码覆盖率为 0
### 下周工作

- 继续完成符号推导工作
- 分析 ci 原因

### 导师点评
Original file line number Diff line number Diff line change
@@ -0,0 +1,32 @@
### 姓名

方国勇

### 实习项目

CINN 符号推导

### 本周工作

1. 尝试在 Windows 上编译 CINN,未果
2. 修复之前发现的 ci 覆盖率为 0 的问题,并修复存量问题
1. https://github.com/PaddlePaddle/Paddle/pull/68607
3. 完善 `det`, `is_empty`,`eye` 的接口实现
1. https://github.com/PaddlePaddle/Paddle/pull/68547
2. https://github.com/PaddlePaddle/Paddle/pull/68548
4. 提交 `prune_gate_by_capacity`, `frame` 的符号推导接口
1. https://github.com/PaddlePaddle/Paddle/pull/68644
2. 学习了 OpTest 的单测写法,给 `prune_gate_by_capacity` 新增了一个 Optest 单测
5. 在 4 的基础上,参考 OpTest 的符号检查,尝试给 unittest 添加符号推导检查的方法
1. https://github.com/PaddlePaddle/Paddle/pull/68670
2. 使用 Value 而不是 fetch_op 简化了检查写法
6. 优化 PIR 下检查 view tensor 被 inplace api 使用时检查和报错优化
1. https://github.com/PaddlePaddle/Paddle/pull/68669

### 下周工作

1. 完善已提交 pr
2. 完成中等以及复杂的算子推导接口


### 导师点评
Original file line number Diff line number Diff line change
@@ -0,0 +1,37 @@
### 姓名

方国勇

### 实习项目

CINN 符号推导

### 本周工作

1. 修复 prune_gate_by_capacity 的编译问题 https://github.com/PaddlePaddle/Paddle/pull/68644
2. 新增 random_routing 符号推导接口 https://github.com/PaddlePaddle/Paddle/pull/68670
1. 没有单测
3. 关闭部分单测
1. https://github.com/PaddlePaddle/Paddle/pull/68841
2. https://github.com/PaddlePaddle/Paddle/pull/68851
3. https://github.com/PaddlePaddle/Paddle/pull/68853
4. 新增 LSTM 符号推导
1. 找到的 OpTest 单测,包含 LOD 信息,过于古老

### 下周工作

1. 本周完成已提交 pr 的修复和合入
| pr | 类型 | 备注 |
| --- | --- | --- |
| https://github.com/PaddlePaddle/Paddle/pull/68907 | LSTM | 新提交的 pr|
| https://github.com/PaddlePaddle/Paddle/pull/68670 | RandomRouting | 按照 review 意见修改完成|
| https://github.com/PaddlePaddle/Paddle/pull/68644 | frame prune_gate_by_capacity | 按照 review 意见修改|
| https://github.com/PaddlePaddle/Paddle/pull/68548 | eye | ci 未过|
| https://github.com/PaddlePaddle/Paddle/pull/68547 | is_empty det | 已 approve, ci 未过|
| https://github.com/PaddlePaddle/Paddle/pull/68841 | 关闭检查 | 已 approve |
| https://github.com/PaddlePaddle/Paddle/pull/68851 | 关闭检查 | 按照 review 意见修改完成 |
|.https://github.com/PaddlePaddle/Paddle/pull/68853 | 关闭检查 |已 approve |
2. 提交 lstsq, llm_int8_linear, instance_norm 的接口实现 pr
1. 每天可以新增一个接口(涉及 kernel 的可能开始会比较慢)和对以往 pr 的修复

### 导师点评
Original file line number Diff line number Diff line change
@@ -0,0 +1,31 @@
### 姓名

方国勇

### 实习项目

CINN 符号推导

### 本周工作

- [LSTM](https://github.com/PaddlePaddle/Paddle/pull/68907)
- [lstsq](https://github.com/PaddlePaddle/Paddle/pull/68917)
- [llm_int8_linear](https://github.com/PaddlePaddle/Paddle/pull/68929)
- [assign_pos](https://github.com/PaddlePaddle/Paddle/pull/68947)
- [instance_norm](https://github.com/PaddlePaddle/Paddle/pull/68964)
- [matrix_rank_tol](https://github.com/PaddlePaddle/Paddle/pull/68975) 需要检查工具函数实现
- [shuffle_batch](https://github.com/PaddlePaddle/Paddle/pull/68978)
- [weight_{dequantize.quantize}](https://github.com/PaddlePaddle/Paddle/pull/68979)
- [uniform_random_batch_size_like](https://github.com/PaddlePaddle/Paddle/pull/68980) 存在 GetOrCreateExprVecFromData 的情况
- [tensor_unfold](https://github.com/PaddlePaddle/Paddle/pull/68981)
- [flash_attn_qkvpacked](https://github.com/PaddlePaddle/Paddle/pull/68984) 需要判断是否可重用
- [array_read](https://github.com/PaddlePaddle/Paddle/pull/69017) manual_op.cc 中


### 下周工作

1. 完善已提交 pr
2. 提交一个防冲突的 pr 占位
3. addnarray 可能需要新的 ShapeOrDataDims 吗, `using RankTensorArrryListShape = std::vector<RankTensorArrayShape>`

### 导师点评
Original file line number Diff line number Diff line change
@@ -0,0 +1,44 @@
### 姓名

方国勇

### 实习项目

CINN 符号推导

### 本周工作

#### 合入 pr
- https://github.com/PaddlePaddle/Paddle/pull/68984
- https://github.com/PaddlePaddle/Paddle/pull/68964
- https://github.com/PaddlePaddle/Paddle/pull/68929
- https://github.com/PaddlePaddle/Paddle/pull/68917
- https://github.com/PaddlePaddle/Paddle/pull/68907

#### 完善 pr
- [frame, prune_gate_by_capacity](https://github.com/PaddlePaddle/Paddle/pull/68644)
- [same_operand_with_result.cc](https://github.com/PaddlePaddle/Paddle/pull/68841)
- [assign_ops](https://github.com/PaddlePaddle/Paddle/pull/68947)
- [matrix_rank_tol](https://github.com/PaddlePaddle/Paddle/pull/68975)
- [shuffle_batch](https://github.com/PaddlePaddle/Paddle/pull/68978)
- [weight_{dequantize.quantize}](https://github.com/PaddlePaddle/Paddle/pull/68979)
- [uniform_random_batch_size_like](https://github.com/PaddlePaddle/Paddle/pull/68980)
- [tensor_unfold](https://github.com/PaddlePaddle/Paddle/pull/68981)
- [array_read](https://github.com/PaddlePaddle/Paddle/pull/69017)

#### 新增 PR
- [create_array_like&&has_elements](https://github.com/PaddlePaddle/Paddle/pull/69055) manual op
- [tensor_to_array&&select_output](https://github.com/PaddlePaddle/Paddle/pull/69055) manual op
- [FloorDivide](https://github.com/PaddlePaddle/Paddle/pull/69167) SIR_32

#### Others

- 0D Tensor,在有 data 时,data.size() = 1 , shape = {}
- data 区数据在 Tensor 作为 shape 的时候,以及可以简单运算或变换得到 1D Tensor 的时候,可以添加 data 区。

### 下周工作

1. 修复 SIR_32 流水线问题
2.

### 导师点评
Loading