| .. |
|
all_reduce
|
156f577f79
feat: switch from `PYBIND11_MODULE` to `TORCH_LIBRARY` (#569)
|
пре 7 месеци |
|
attention
|
156f577f79
feat: switch from `PYBIND11_MODULE` to `TORCH_LIBRARY` (#569)
|
пре 7 месеци |
|
backup
|
f8dfac6372
chore: attention refactor and upstream sync apr01 (#365)
|
пре 11 месеци |
|
cpu
|
271a680026
feat: inference support for PowerPC ISA
|
пре 7 месеци |
|
hadamard
|
5d288aa76c
feat: add fast hadamard transformation kernels (#232)
|
пре 1 година |
|
moe
|
156f577f79
feat: switch from `PYBIND11_MODULE` to `TORCH_LIBRARY` (#569)
|
пре 7 месеци |
|
punica
|
4f42985b5c
feat: qwen2 lora shapes
|
пре 7 месеци |
|
quantization
|
b03b4d4c8c
fix: compute cutlass 3.x epilogues in fp32 instead of 16
|
пре 7 месеци |
|
activation_kernels.cu
|
c0c336aaa3
refactor: registry for processing model inputs; quick_gelu; clip model support
|
пре 7 месеци |
|
cache.h
|
156f577f79
feat: switch from `PYBIND11_MODULE` to `TORCH_LIBRARY` (#569)
|
пре 7 месеци |
|
cache_kernels.cu
|
156f577f79
feat: switch from `PYBIND11_MODULE` to `TORCH_LIBRARY` (#569)
|
пре 7 месеци |
|
cuda_compat.h
|
00acf371f9
rocm: fused topk softmax
|
пре 7 месеци |
|
cuda_utils.h
|
156f577f79
feat: switch from `PYBIND11_MODULE` to `TORCH_LIBRARY` (#569)
|
пре 7 месеци |
|
cuda_utils_kernels.cu
|
156f577f79
feat: switch from `PYBIND11_MODULE` to `TORCH_LIBRARY` (#569)
|
пре 7 месеци |
|
dispatch_utils.h
|
156f577f79
feat: switch from `PYBIND11_MODULE` to `TORCH_LIBRARY` (#569)
|
пре 7 месеци |
|
layernorm_kernels.cu
|
156f577f79
feat: switch from `PYBIND11_MODULE` to `TORCH_LIBRARY` (#569)
|
пре 7 месеци |
|
ops.h
|
c0c336aaa3
refactor: registry for processing model inputs; quick_gelu; clip model support
|
пре 7 месеци |
|
pos_encoding_kernels.cu
|
156f577f79
feat: switch from `PYBIND11_MODULE` to `TORCH_LIBRARY` (#569)
|
пре 7 месеци |
|
reduction.cuh
|
aba03b4756
feat: dynamic per-token activation quantization
|
пре 7 месеци |
|
registration.h
|
156f577f79
feat: switch from `PYBIND11_MODULE` to `TORCH_LIBRARY` (#569)
|
пре 7 месеци |
|
torch_bindings.cpp
|
c0c336aaa3
refactor: registry for processing model inputs; quick_gelu; clip model support
|
пре 7 месеци |
AlpinDale
b03b4d4c8c
fix: compute cutlass 3.x epilogues in fp32 instead of 16