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.

Sign up for GitHub

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

Jump to bottom

Dev #74

Merged
merged 64 commits into from
Feb 19, 2024
Merged

Dev #74

merged 64 commits into from
Feb 19, 2024

Conversation

YdrMaster
Copy link
Collaborator

No description provided.

@YdrMaster YdrMaster added kernel Add or modify kernel implementation operator Add or modify frontend/computation operator labels Jan 11, 2024
@YdrMaster YdrMaster self-assigned this Jan 11, 2024
This was linked to issues Jan 11, 2024
Mod #66
Closed
Erf #65
Closed
PanZezhong1725 and others added 12 commits January 17, 2024 11:38
@PanZezhong1725
fix (frontend): 修改get输入输出，内存大小，增加compiler设置输入接口
eb4fc4c
@YdrMaster
Merge pull request #77 from InfiniTensor/front_fix
54ffa1a 
fix (frontend): 修改get输入输出，内存大小，增加compiler设置输入接口
@zhangyue207 @YdrMaster
add head file
e76c1fc
@YdrMaster
Merge pull request #79 from InfiniTensor/add-headfile
8b6333d 
Add headfile
@bitzyz
feat: add HardSigmoid cpu/cuda kernel
a7ed032
@bitzyz
feat: add hardswish cpu/cuda kernel
3998833
@kilinchange @bitzyz
fix: fix mod kernel
1ce6d81
@YdrMaster
Merge pull request #78 from InfiniTensor/dev-hardsigmoid
9e4789c 
add HardSigmoid cpu/cuda kernel
@YdrMaster
style: 整理 hardswish 计算
afdfe93 
Signed-off-by: YdrMaster <[email protected]>
@YdrMaster
Merge pull request #73 from InfiniTensor/dev-hardswish
a163eee 
feat: add hardswish cpu/cuda kernel
@YdrMaster
feat(python_ffi): 添加一个字符型的 trace 格式
ec39fd7 
Signed-off-by: YdrMaster <[email protected]>
@YdrMaster
fix(kernel): 双目运算不能交换
20788a3 
Signed-off-by: YdrMaster <[email protected]>
@YdrMaster YdrMaster linked an issue Jan 19, 2024 that may be closed by this pull request
Hardswish #67
Closed
7 tasks
@YdrMaster YdrMaster force-pushed the dev branch 2 times, most recently from 80f7665 to 9acbb3c Compare January 24, 2024 09:54
This was linked to issues Jan 26, 2024
Pad #64
Closed
RMS Normalization #83
Closed
YdrMaster and others added 23 commits January 29, 2024 09:20
@YdrMaster
feat(llm): 注册 rms normalization
add61cb 
Signed-off-by: YdrMaster <[email protected]>
@YdrMaster
feat: 搭建 Attention 在各层的基本结构
daeac28 
Signed-off-by: YdrMaster <[email protected]>
@YdrMaster
feat(llm): 实现 attention 形状推导
722e6d9 
Signed-off-by: YdrMaster <[email protected]>
@YdrMaster
fix(hardware): typo
12e264a 
Signed-off-by: YdrMaster <[email protected]>
@YdrMaster
refactor(kernel): 重构 transpose info 计算
8595be4 
Signed-off-by: YdrMaster <[email protected]>
@YdrMaster
refactor(kernel): 重构 transpose info 和 kernel 的构造
f4d063a 
Signed-off-by: YdrMaster <[email protected]>
@YdrMaster
refactor(kernel): transpose 支持合并连续访存
bbed31e 
Signed-off-by: YdrMaster <[email protected]>
@YdrMaster
feat: 添加 attention cuda
f1faf3e 
Signed-off-by: YdrMaster <[email protected]>
@YdrMaster
build(kernel): 添加 cublasLt
9495516 
Signed-off-by: YdrMaster <[email protected]>
@YdrMaster
fix(kernel): 为 Gather 支持负的 indices 值
6630866 
Signed-off-by: YdrMaster <[email protected]>
@kilinchange
add max/min cuda kernel and test
e237349
@YdrMaster
Merge pull request #84 from InfiniTensor/add_max_min_kernel
35dc6c8 
add max/min kernel
@YdrMaster
fix(frontend): 填充边信息时不能接受输入边不存在
27a8ad6 
Signed-off-by: YdrMaster <[email protected]>
@bitzyz
fix(fronted): 为适配torchvision模型，修改前端从而接调用cpu kernel
eea2249
@bitzyz
fix(fronted): 修复reduce算子对reducesum的opset特判bug
3f5b0ee
@bitzyz
fix(kernel):修复transpose info存在的bug
7db54c1
@YdrMaster
fix(kernel): 为 TransposeInfo 应对更多 coner case 并增加一个单元测试
b3e89a6 
Signed-off-by: YdrMaster <[email protected]>
@YdrMaster
Merge pull request #88 from InfiniTensor/fix-torchvision-models
d076c20 
Fix for apply torchvision models
@bitzyz
feat(kernel): 添加exp算子
3199eb5
@YdrMaster
fix(kernel): 没有引用的输出张量也要分配，但计算完后立即释放
e3febd9 
Signed-off-by: YdrMaster <[email protected]>
@PanZezhong1725
feat (dist): nccl通信库接入，allreduce算子
16870d6
@YdrMaster
Merge pull request #92 from InfiniTensor/dist-merge
34ed834 
feat (dist): nccl通信库接入，allreduce算子
@YdrMaster
Merge pull request #91 from InfiniTensor/dev-exp
3cb6f5d 
feat(kernel): 添加exp算子
@YdrMaster YdrMaster merged commit a813939 into master Feb 19, 2024
4 checks passed
@YdrMaster YdrMaster deleted the dev branch February 19, 2024 10:18
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers
No reviews
Assignees

@YdrMaster YdrMaster

Labels
kernel Add or modify kernel implementation operator Add or modify frontend/computation operator
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.

RMS Normalization Hardswish Mod Erf Pad
6 participants
@YdrMaster @kilinchange @PanZezhong1725 @zhangyue207 @bitzyz @Chamberlain0w0