Merge pull request #5310 from hpcaitech/feature/npu

Feature/npu

Merge pull request #5310 from hpcaitech/feature/npu
Feature/npu
8823cc48 · Frank Lee · GitHub · bce9499e · 73f4dc57 · bce9499e
Unverified Commit 8823cc48 authored Jan 29, 2024 by Frank Lee Committed by GitHub Jan 29, 2024
20 changed files
--- a/colossalai/kernel/cuda_native/csrc/kernels/include/strided_batch_gemm.h
+++ b/colossalai/kernel/cuda_native/csrc/kernels/include/strided_batch_gemm.h
--- a/colossalai/kernel/cuda_native/csrc/kernels/normalize_kernels.cu
+++ b/colossalai/kernel/cuda_native/csrc/kernels/normalize_kernels.cu
--- a/colossalai/kernel/cuda_native/csrc/kernels/softmax_kernels.cu
+++ b/colossalai/kernel/cuda_native/csrc/kernels/softmax_kernels.cu
--- a/colossalai/kernel/cuda_native/csrc/kernels/transform_kernels.cu
+++ b/colossalai/kernel/cuda_native/csrc/kernels/transform_kernels.cu
--- a/colossalai/kernel/cuda_native/csrc/multihead_attention_1d.cpp
+++ b/colossalai/kernel/cuda_native/csrc/multihead_attention_1d.cpp
--- a/colossalai/kernel/cuda_native/csrc/multihead_attention_1d.h
+++ b/colossalai/kernel/cuda_native/csrc/multihead_attention_1d.h
--- a/colossalai/kernel/cuda_native/csrc/smoothquant/binding.cpp
+++ b/colossalai/kernel/cuda_native/csrc/smoothquant/binding.cpp
--- a/colossalai/kernel/cuda_native/csrc/smoothquant/linear.cu
+++ b/colossalai/kernel/cuda_native/csrc/smoothquant/linear.cu
--- a/colossalai/kernel/cuda_native/csrc/smoothquant/linear.h
+++ b/colossalai/kernel/cuda_native/csrc/smoothquant/linear.h
--- a/colossalai/kernel/cuda_native/mha/__init__.py
+++ b/colossalai/kernel/cuda_native/mha/__init__.py
--- a/colossalai/kernel/cuda_native/mha/flash_attn_2.py
+++ b/colossalai/kernel/cuda_native/mha/flash_attn_2.py
--- a/colossalai/kernel/cuda_native/mha/mem_eff_attn.py
+++ b/colossalai/kernel/cuda_native/mha/mem_eff_attn.py
--- a/colossalai/kernel/cuda_native/mha/utils.py
+++ b/colossalai/kernel/cuda_native/mha/utils.py
--- a/colossalai/kernel/cuda_native/multihead_attention.py
+++ b/colossalai/kernel/cuda_native/multihead_attention.py
--- a/colossalai/kernel/extensions
+++ b/colossalai/kernel/extensions
+../../extensions
\ No newline at end of file
--- a/colossalai/kernel/jit/option.py
+++ b/colossalai/kernel/jit/option.py
--- a/colossalai/kernel/kernel_loader.py
+++ b/colossalai/kernel/kernel_loader.py
--- a/colossalai/kernel/op_builder
+++ b/colossalai/kernel/op_builder
--- a/colossalai/legacy/amp/naive_amp/_fp16_optimizer.py
+++ b/colossalai/legacy/amp/naive_amp/_fp16_optimizer.py
--- a/colossalai/legacy/amp/torch_amp/torch_amp.py
+++ b/colossalai/legacy/amp/torch_amp/torch_amp.py