| .. |
|
all_reduce
|
6305e6f3f2
fix: no repeated IPC registration (#227)
|
1 жил өмнө |
|
attention
|
8e1cd54497
fix: do not include fp8 for rocm (#271)
|
1 жил өмнө |
|
hadamard
|
5d288aa76c
feat: add fast hadamard transformation kernels (#232)
|
1 жил өмнө |
|
moe
|
7d6ba53602
feat: fused top-k kernels for MoE (#273)
|
1 жил өмнө |
|
punica
|
c0aac15421
feat: S-LoRA support (#222)
|
1 жил өмнө |
|
quantization
|
82955ba440
fix: backport bnb kernels (#297)
|
1 жил өмнө |
|
activation_kernels.cu
|
e31c6f0b45
feat: refactor modeling logic and support more models (#274)
|
1 жил өмнө |
|
cache.h
|
31c95011a6
feat: FP8 E5M2 KV Cache (#226)
|
1 жил өмнө |
|
cache_kernels.cu
|
8e1cd54497
fix: do not include fp8 for rocm (#271)
|
1 жил өмнө |
|
cuda_compat.h
|
8fa608aeb7
feat: replace Ray with NCCL for control plane comms (#221)
|
1 жил өмнө |
|
cuda_utils.h
|
31c95011a6
feat: FP8 E5M2 KV Cache (#226)
|
1 жил өмнө |
|
cuda_utils_kernels.cu
|
31c95011a6
feat: FP8 E5M2 KV Cache (#226)
|
1 жил өмнө |
|
dispatch_utils.h
|
d9b65e6c5f
feat: DeepSeek MoE support (#237)
|
1 жил өмнө |
|
layernorm_kernels.cu
|
8fa608aeb7
feat: replace Ray with NCCL for control plane comms (#221)
|
1 жил өмнө |
|
ops.h
|
e0c35bb353
feat: bitsandbytes and `--load-in{4,8}bit` support (#294)
|
1 жил өмнө |
|
pos_encoding_kernels.cu
|
8fa608aeb7
feat: replace Ray with NCCL for control plane comms (#221)
|
1 жил өмнө |
|
pybind.cpp
|
e0c35bb353
feat: bitsandbytes and `--load-in{4,8}bit` support (#294)
|
1 жил өмнө |
|
reduction.cuh
|
2755a48d51
merge dev branch into main (#153)
|
1 жил өмнө |
AlpinDale
82955ba440
fix: backport bnb kernels (#297)