AlpinDale ba371fbbbd feat: AWQ marlin kernels (#603) преди 5 месеца
..
all_reduce 156f577f79 feat: switch from `PYBIND11_MODULE` to `TORCH_LIBRARY` (#569) преди 5 месеца
attention 9d7beaa5b9 chore: separate kv_scale into k_scale and v_scale преди 5 месеца
backup f8dfac6372 chore: attention refactor and upstream sync apr01 (#365) преди 9 месеца
cpu 9d7beaa5b9 chore: separate kv_scale into k_scale and v_scale преди 5 месеца
hadamard 5d288aa76c feat: add fast hadamard transformation kernels (#232) преди 11 месеца
mamba f5d52320da Port mamba kernels to Aphrodite (#595) преди 5 месеца
moe 156f577f79 feat: switch from `PYBIND11_MODULE` to `TORCH_LIBRARY` (#569) преди 5 месеца
prepare_inputs dd18c5042c move prepare_inputs to the GPU (#596) преди 5 месеца
punica 4f42985b5c feat: qwen2 lora shapes преди 5 месеца
quantization ba371fbbbd feat: AWQ marlin kernels (#603) преди 5 месеца
activation_kernels.cu c0c336aaa3 refactor: registry for processing model inputs; quick_gelu; clip model support преди 5 месеца
cache.h 9d7beaa5b9 chore: separate kv_scale into k_scale and v_scale преди 5 месеца
cache_kernels.cu 9d7beaa5b9 chore: separate kv_scale into k_scale and v_scale преди 5 месеца
cuda_compat.h 00acf371f9 rocm: fused topk softmax преди 5 месеца
cuda_utils.h 156f577f79 feat: switch from `PYBIND11_MODULE` to `TORCH_LIBRARY` (#569) преди 5 месеца
cuda_utils_kernels.cu 156f577f79 feat: switch from `PYBIND11_MODULE` to `TORCH_LIBRARY` (#569) преди 5 месеца
dispatch_utils.h 156f577f79 feat: switch from `PYBIND11_MODULE` to `TORCH_LIBRARY` (#569) преди 5 месеца
layernorm_kernels.cu 156f577f79 feat: switch from `PYBIND11_MODULE` to `TORCH_LIBRARY` (#569) преди 5 месеца
ops.h dd18c5042c move prepare_inputs to the GPU (#596) преди 5 месеца
pos_encoding_kernels.cu 156f577f79 feat: switch from `PYBIND11_MODULE` to `TORCH_LIBRARY` (#569) преди 5 месеца
reduction.cuh aba03b4756 feat: dynamic per-token activation quantization преди 5 месеца
registration.h 156f577f79 feat: switch from `PYBIND11_MODULE` to `TORCH_LIBRARY` (#569) преди 5 месеца
torch_bindings.cpp ba371fbbbd feat: AWQ marlin kernels (#603) преди 5 месеца