Star451
259
代码
代码
Issues35
Pull Requests86
Actions
项目讨论
Wiki
分析
Star451
259
文件最后提交记录最后更新时间
copy_gm_to_l1.hpp
49_ascend950_flash_attention_infer torch_catlass框架接口补齐 Co-authored-by: CheaterAbec<niyuhao2@huawei.com> # message auto-generated for no-merge-commit merge: !684 merge master into master 49_ascend950_flash_attention_infer torch_catlass框架接口补齐 Created-by: CheaterAbec Commit-by: CheaterAbec Merged-by: cann-robot Description: ## 描述 <!--在这里详细描述你的改动，包括改动的原因和所采取的方法。--> 49_ascend950_flash_attention_infer torch_catlass框架接口补齐 ## 关联的Issue <!-- 如果这个PR是为了解决特定的Issue，请在这里提供Issue链接。--> ## 原因 <!--说明此次改动的目的、解决的问题等，应与类型标签匹配 --> ## 测试 <!--描述进行了哪些测试来验证你的改动。包括但不限于二级冒烟、算子泛化等。--> cmd ======================================================================================================================= test session starts ======================================================================================================================= platform linux -- Python 3.11.15, pytest-9.0.3, pluggy-1.6.0 -- /home/miniconda3/envs/niyuhao/bin/python3 cachedir: .pytest_cache rootdir: /home/n00955621/catlass_630/catlass_950/tests/optest configfile: pyproject.toml collected 20 items tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_masked_b2_fp16] [W609 12:41:41.579147854 TensorFactories.cpp:340] Warning: Cannot create tensor with interal format while allow_internel_format=False, tensor will be created with base format. (function operator()) PASSED tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_nomask_b2_fp16] PASSED tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_masked_gqa_fp16] PASSED tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_masked_readme_shape] PASSED tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_masked_decode_fp16] PASSED tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_masked_head64_fp16] [W609 12:41:41.191383901 ToKernelNpu.cpp:41] Warning: Device do not support double dtype now, dtype cast replace with float. (function operator()) XFAIL (native example 49 fails head_dim=64 (Compare failed); kernel template targets embed=128) tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_masked_b2_bf16] PASSED tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_masked_b1_fp16] PASSED tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_masked_b4_fp16] PASSED tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_nomask_b1_fp16] PASSED tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_masked_sq128_fp16] PASSED tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_masked_long_kv_fp16] PASSED tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_masked_small_fp16] PASSED tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_masked_gqa8x4_fp16] PASSED tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_masked_mha_fp16] PASSED tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_nomask_gqa_fp16] PASSED tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_masked_decode_bf16] PASSED tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_nomask_decode_fp16] PASSED tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_masked_b2_decode_fp16] PASSED tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_nomask_readme_bf16] PASSED ================================================================================================================== 19 passed, 1 xfailed in 9.63s ================================================================================================================== ## 文档更新 <!--如果这个PR包含文档的更新，请在这里指出。例如：更新了README.md文件。--> ## 类型标签 <!-- [x] 表示选中 --> - [ ] Bug修复 - [ ] 新特性 - [ ] 性能优化 - [ ] 文档更新 - [ ] 其他，请描述： See merge request: cann/catlass!684
2 个月前
copy_l0c_to_dst.hpp
fix CopyL0CToDstQuantMode Co-authored-by: chen_xu0513<chenxu210@huawei.com> # message auto-generated for no-merge-commit merge: !510 merge master into master fix CopyL0CToDstQuantMode Created-by: chen_xu0513 Commit-by: chen_xu0513 Merged-by: cann-robot Description: ## 描述 <!--在这里详细描述你的改动，包括改动的原因和所采取的方法。--> ## 关联的Issue <!-- 如果这个PR是为了解决特定的Issue，请在这里提供Issue链接。--> ## 原因 CopyL0CToDstQuantMode引入编译错误<!--说明此次改动的目的、解决的问题等，应与类型标签匹配 --> ## 测试 <!--描述进行了哪些测试来验证你的改动。包括但不限于二级冒烟、算子泛化等。--> ## 文档更新 <!--如果这个PR包含文档的更新，请在这里指出。例如：更新了README.md文件。--> ## 类型标签 <!-- [x] 表示选中 --> - [x] Bug修复 - [ ] 新特性 - [ ] 性能优化 - [ ] 文档更新 - [ ] 其他，请描述： See merge request: cann/catlass!510
4 个月前
copy_l0c_to_gm.hpp
文档低错修改 Co-authored-by: init__zhb__<zhanghaobo6@huawei.com> # message auto-generated for no-merge-commit merge: !597 merge br.docup into master 文档低错修改 Created-by: init__zhb__ Commit-by: init__zhb__ Merged-by: cann-robot Description: ## 描述 修改/docs目录下相关文档，减少事实错误，同时修复header guard不规范问题，以及一些杂项修改（copyright缺失等） ## 关联的Issue <!-- 如果这个PR是为了解决特定的Issue，请在这里提供Issue链接。--> ## 原因 <!--说明此次改动的目的、解决的问题等，应与类型标签匹配 --> ## 测试 <!--描述进行了哪些测试来验证你的改动。包括但不限于二级冒烟、算子泛化等。--> ## 文档更新 <!--如果这个PR包含文档的更新，请在这里指出。例如：更新了README.md文件。--> ## 类型标签 <!-- [x] 表示选中 --> - [ ] Bug修复 - [ ] 新特性 - [ ] 性能优化 - [ ] 文档更新 - [ ] 其他，请描述： See merge request: cann/catlass!597
3 个月前
copy_l0c_to_ub.hpp
chunkprefill_mask_pa Co-authored-by: lccchh<linchenghao@huawei.com> # message auto-generated for no-merge-commit merge: !700 merge 950_ChunkPrefill into master chunkprefill_mask_pa Created-by: lccchh Commit-by: lccchh Merged-by: cann-robot Description: ## 描述 新增适配Ascend950硬件的，不同于库上存量模板，使用[2048, 2048]压缩mask的flash attention，并支持以下特性： ![image.png](https://raw.gitcode.com/user-images/assets/7631999/7719ab83-33aa-45c7-b276-4b81a1b94833/image.png 'image.png') ## 关联的Issue https://gitcode.com/cann/catlass/issues/234 ## 原因 <!--说明此次改动的目的、解决的问题等，应与类型标签匹配 --> ## 测试 <!--描述进行了哪些测试来验证你的改动。包括但不限于二级冒烟、算子泛化等。--> ## 文档更新 <!--如果这个PR包含文档的更新，请在这里指出。例如：更新了README.md文件。--> ## 类型标签 <!-- [x] 表示选中 --> - [ ] Bug修复 - [ ] 新特性 - [ ] 性能优化 - [ ] 文档更新 - [ ] 其他，请描述： See merge request: cann/catlass!700
1 个月前
copy_l1_to_bt.hpp
文档低错修改 Co-authored-by: init__zhb__<zhanghaobo6@huawei.com> # message auto-generated for no-merge-commit merge: !597 merge br.docup into master 文档低错修改 Created-by: init__zhb__ Commit-by: init__zhb__ Merged-by: cann-robot Description: ## 描述 修改/docs目录下相关文档，减少事实错误，同时修复header guard不规范问题，以及一些杂项修改（copyright缺失等） ## 关联的Issue <!-- 如果这个PR是为了解决特定的Issue，请在这里提供Issue链接。--> ## 原因 <!--说明此次改动的目的、解决的问题等，应与类型标签匹配 --> ## 测试 <!--描述进行了哪些测试来验证你的改动。包括但不限于二级冒烟、算子泛化等。--> ## 文档更新 <!--如果这个PR包含文档的更新，请在这里指出。例如：更新了README.md文件。--> ## 类型标签 <!-- [x] 表示选中 --> - [ ] Bug修复 - [ ] 新特性 - [ ] 性能优化 - [ ] 文档更新 - [ ] 其他，请描述： See merge request: cann/catlass!597
3 个月前
copy_l1_to_l0a.hpp
49_ascend950_flash_attention_infer torch_catlass框架接口补齐 Co-authored-by: CheaterAbec<niyuhao2@huawei.com> # message auto-generated for no-merge-commit merge: !684 merge master into master 49_ascend950_flash_attention_infer torch_catlass框架接口补齐 Created-by: CheaterAbec Commit-by: CheaterAbec Merged-by: cann-robot Description: ## 描述 <!--在这里详细描述你的改动，包括改动的原因和所采取的方法。--> 49_ascend950_flash_attention_infer torch_catlass框架接口补齐 ## 关联的Issue <!-- 如果这个PR是为了解决特定的Issue，请在这里提供Issue链接。--> ## 原因 <!--说明此次改动的目的、解决的问题等，应与类型标签匹配 --> ## 测试 <!--描述进行了哪些测试来验证你的改动。包括但不限于二级冒烟、算子泛化等。--> cmd ======================================================================================================================= test session starts ======================================================================================================================= platform linux -- Python 3.11.15, pytest-9.0.3, pluggy-1.6.0 -- /home/miniconda3/envs/niyuhao/bin/python3 cachedir: .pytest_cache rootdir: /home/n00955621/catlass_630/catlass_950/tests/optest configfile: pyproject.toml collected 20 items tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_masked_b2_fp16] [W609 12:41:41.579147854 TensorFactories.cpp:340] Warning: Cannot create tensor with interal format while allow_internel_format=False, tensor will be created with base format. (function operator()) PASSED tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_nomask_b2_fp16] PASSED tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_masked_gqa_fp16] PASSED tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_masked_readme_shape] PASSED tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_masked_decode_fp16] PASSED tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_masked_head64_fp16] [W609 12:41:41.191383901 ToKernelNpu.cpp:41] Warning: Device do not support double dtype now, dtype cast replace with float. (function operator()) XFAIL (native example 49 fails head_dim=64 (Compare failed); kernel template targets embed=128) tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_masked_b2_bf16] PASSED tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_masked_b1_fp16] PASSED tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_masked_b4_fp16] PASSED tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_nomask_b1_fp16] PASSED tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_masked_sq128_fp16] PASSED tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_masked_long_kv_fp16] PASSED tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_masked_small_fp16] PASSED tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_masked_gqa8x4_fp16] PASSED tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_masked_mha_fp16] PASSED tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_nomask_gqa_fp16] PASSED tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_masked_decode_bf16] PASSED tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_nomask_decode_fp16] PASSED tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_masked_b2_decode_fp16] PASSED tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_nomask_readme_bf16] PASSED ================================================================================================================== 19 passed, 1 xfailed in 9.63s ================================================================================================================== ## 文档更新 <!--如果这个PR包含文档的更新，请在这里指出。例如：更新了README.md文件。--> ## 类型标签 <!-- [x] 表示选中 --> - [ ] Bug修复 - [ ] 新特性 - [ ] 性能优化 - [ ] 文档更新 - [ ] 其他，请描述： See merge request: cann/catlass!684
2 个月前
copy_l1_to_l0b.hpp
49_ascend950_flash_attention_infer torch_catlass框架接口补齐 Co-authored-by: CheaterAbec<niyuhao2@huawei.com> # message auto-generated for no-merge-commit merge: !684 merge master into master 49_ascend950_flash_attention_infer torch_catlass框架接口补齐 Created-by: CheaterAbec Commit-by: CheaterAbec Merged-by: cann-robot Description: ## 描述 <!--在这里详细描述你的改动，包括改动的原因和所采取的方法。--> 49_ascend950_flash_attention_infer torch_catlass框架接口补齐 ## 关联的Issue <!-- 如果这个PR是为了解决特定的Issue，请在这里提供Issue链接。--> ## 原因 <!--说明此次改动的目的、解决的问题等，应与类型标签匹配 --> ## 测试 <!--描述进行了哪些测试来验证你的改动。包括但不限于二级冒烟、算子泛化等。--> cmd ======================================================================================================================= test session starts ======================================================================================================================= platform linux -- Python 3.11.15, pytest-9.0.3, pluggy-1.6.0 -- /home/miniconda3/envs/niyuhao/bin/python3 cachedir: .pytest_cache rootdir: /home/n00955621/catlass_630/catlass_950/tests/optest configfile: pyproject.toml collected 20 items tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_masked_b2_fp16] [W609 12:41:41.579147854 TensorFactories.cpp:340] Warning: Cannot create tensor with interal format while allow_internel_format=False, tensor will be created with base format. (function operator()) PASSED tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_nomask_b2_fp16] PASSED tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_masked_gqa_fp16] PASSED tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_masked_readme_shape] PASSED tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_masked_decode_fp16] PASSED tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_masked_head64_fp16] [W609 12:41:41.191383901 ToKernelNpu.cpp:41] Warning: Device do not support double dtype now, dtype cast replace with float. (function operator()) XFAIL (native example 49 fails head_dim=64 (Compare failed); kernel template targets embed=128) tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_masked_b2_bf16] PASSED tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_masked_b1_fp16] PASSED tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_masked_b4_fp16] PASSED tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_nomask_b1_fp16] PASSED tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_masked_sq128_fp16] PASSED tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_masked_long_kv_fp16] PASSED tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_masked_small_fp16] PASSED tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_masked_gqa8x4_fp16] PASSED tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_masked_mha_fp16] PASSED tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_nomask_gqa_fp16] PASSED tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_masked_decode_bf16] PASSED tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_nomask_decode_fp16] PASSED tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_masked_b2_decode_fp16] PASSED tests/test_49_ascend950_flash_attention_infer.py::test_ascend950_flash_attention_infer[paged_nomask_readme_bf16] PASSED ================================================================================================================== 19 passed, 1 xfailed in 9.63s ================================================================================================================== ## 文档更新 <!--如果这个PR包含文档的更新，请在这里指出。例如：更新了README.md文件。--> ## 类型标签 <!-- [x] 表示选中 --> - [ ] Bug修复 - [ ] 新特性 - [ ] 性能优化 - [ ] 文档更新 - [ ] 其他，请描述： See merge request: cann/catlass!684
2 个月前