ThunderKittens
Tile primitives for speedy kernels
파일 탐색기
최종 버전 다운로드 (.zip)- settings.json
- attn.png
- kittens.png
- thunderkittens.png
- __init__.py
- convolution.py
- linear_attention.py
- rotary.py
- slide_attention.py
- __init__.py
- layer_norm.py
- __init__.py
- benchmark-lin-attn-fwd-B16.png
- benchmark-lin-attn-fwd-B4.png
- benchmark-lin-attn-fwd-L8192.png
- benchmark_input1000_output1.png
- benchmark_input16000_output1.png
- benchmark_input8000_output1.png
- __init__.py
- block.py
- embeddings.py
- gpt.py
- mha.py
- mlp.py
- hf.py
- utils.py
- __init__.py
- generation.py
- document_ie_based.py
- generate_based.py
- README.md
- llama_3.1_8b_distill_config.yaml
- llama_3.1_8b_finetune_config.yaml
- llama_3.1_8b_model_config.yaml
- __init__.py
- pretrained.py
- transformers_modeling_llama.py
- transformers_modeling_utils.py
- __init__.py
- logging.py
- setup.py
- __init__.py
- demo_8b.sh
- demo_llama_hf.py
- __init__.py
- linear_attention.py
- linear_window_attention_tk.py
- linear_window_attention_tk_gen.py
- linear_window_attention_tk_long.py
- utils.py
- __init__.py
- convert_model.py
- feature_map.py
- load_model.py
- modeling_llama.py
- peft.py
- pretrained.py
- rotary.py
- utils.py
- __init__.py
- logging.py
- setup.py
- __init__.py
- demo_8b.sh
- demo_lolcats_hf.py
- __init__.py
- pretrained.py
- transformers_modeling_qwen.py
- transformers_modeling_utils.py
- __init__.py
- logging.py
- setup.py
- __init__.py
- demo_8b.sh
- demo_qwen_hf.py
- requirements.txt
- base_ops.cuh
- base_types.cuh
- common.cuh
- multimem.cuh
- util.cuh
- complex_global_to_register.cuh
- complex_global_to_shared.cuh
- complex_shared_to_register.cuh
- global_to_register.cuh
- global_to_shared.cuh
- parallel_global_to_global.cuh
- shared_to_register.cuh
- tensor_to_register.cuh
- tile.cuh
- tma.cuh
- tma_cluster.cuh
- global_to_register.cuh
- global_to_shared.cuh
- shared_to_register.cuh
- tma.cuh
- tma_cluster.cuh
- vec.cuh
- memory.cuh
- 64x112.impl
- 64x128.impl
- 64x144.impl
- 64x16.impl
- 64x160.impl
- 64x176.impl
- 64x192.impl
- 64x208.impl
- 64x224.impl
- 64x240.impl
- 64x256.impl
- 64x32.impl
- 64x48.impl
- 64x64.impl
- 64x80.impl
- 64x96.impl
- base.cuh
- mma.cuh
- tcgen05.cuh
- warp.cuh
- warpgroup.cuh
- complex_conversions.cuh
- complex_maps.cuh
- conversions.cuh
- maps.cuh
- reductions.cuh
- tile.cuh
- conversions.cuh
- maps.cuh
- reductions.cuh
- vec.cuh
- register.cuh
- conversions.cuh
- maps.cuh
- reductions.cuh
- tile.cuh
- conversions.cuh
- maps.cuh
- reductions.cuh
- vec.cuh
- shared.cuh
- sync.cuh
- tma.cuh
- tma_cluster.cuh
- util.cuh
- group.cuh
- shared_to_tensor.cuh
- tile.cuh
- tma.cuh
- tma.cuh
- vec.cuh
- memory.cuh
- mma.cuh
- tcgen05.cuh
- sync.cuh
- tma.cuh
- util.cuh
- thread.cuh
- ops.cuh
- broker.cuh
- club.cuh
- parallel_tensor.cuh
- profiler.cuh
- pyutils.cuh
- torchutils.cuh
- cgl.cuh
- gl.cuh
- global.cuh
- tma.cuh
- util.cuh
- crt.cuh
- crv.cuh
- register.cuh
- rt.cuh
- rt_base.cuh
- rt_layout.cuh
- rv.cuh
- rv_layout.cuh
- cst.cuh
- csv.cuh
- descriptor.cuh
- shared.cuh
- st.cuh
- sv.cuh
- ipc.cuh
- pgl.cuh
- system.cuh
- vmm.cuh
- tensor.cuh
- tt.cuh
- types.cuh
- kittens.cuh
- bf16_b300_mha_causal.cu
- Makefile
- test.py
- bf16_b300_mha_noncausal.cu
- Makefile
- test.py
- benchmark.py
- gentests.py
- harness.impl
- Makefile
- mha_h100.cu
- test_correctness.py
- gentests.py
- Makefile
- mha_h100_lcf.cu
- benchmark.py
- gentests.py
- harness.impl
- linear_attn.cu
- Makefile
- tk_fftconv.py
- benchmark.py
- fftconv_non_pc.cu
- fftconv_pc.cu
- gentests.py
- gentests_1024.py
- harness.impl
- Makefile
- pytorch_ref.py
- test_correctness.py
- benchmark.py
- flux_gate.cu
- flux_gelu.cu
- Makefile
- bf16_cublas_gemm.cu
- Makefile
- bf16_cublas_lt_gemm.cu
- Makefile
- README.md
- fp8_cublas_lt_gemm.cu
- Makefile
- int8_cublas_lt_gemm.cu
- Makefile
- Makefile
- mxfp8_cublas_lt_gemm.cu
- Makefile
- nvfp4_cublas_lt_gemm.cu
- bf16_b200_gemm.cu
- Makefile
- bf16_h100_gemm.cu
- Makefile
- launch.cu
- level_01.cu
- level_02.cu
- level_03.cu
- level_04.cu
- level_05.cu
- level_06.cu
- level_07.cu
- level_08.cu
- level_09.cu
- Makefile
- README.md
- launch.cu
- level_01.cu
- level_02.cu
- level_03.cu
- level_04.cu
- level_05.cu
- level_06.cu
- level_07.cu
- level_08.cu
- Makefile
- README.md
- fp8_b200_gemm.cu
- Makefile
- fp8_h100_gemm.cu
- Makefile
- fp8_h100_gemm_scaled.cu
- Makefile
- visualize.py
- int8_b200_gemm.cu
- Makefile
- int8_h100_gemm.cu
- Makefile
- Makefile
- mxfp8_b200_gemm.cu
- test_gemm.py
- test_quantize.py
- Makefile
- nvfp4_b200_gemm.cu
- test_gemm.py
- test_quantize.py
- Makefile
- nvfp4_b300_gemm.cu
- nvfp4_b300_gemm_optimized.cu
- test_gemm.py
- test_quantize.py
- common.cuh
- benchmark.py
- gentests.py
- harness.impl
- hedgehog.cu
- Makefile
- test_correctness.py
- util.py
- layer_norm_triton.py
- benchmark.py
- gentests.py
- harness.impl
- layernorm.cu
- Makefile
- test_correctness.py
- gentests.py
- linear_attention.cu
- Makefile
- ssd_minimal.py
- benchmark.py
- gentests.py
- harness.impl
- harness2.impl
- harness3.impl
- Makefile
- mamba2.cu
- test_correctness.py
- ag_gemm_b200.cu
- ag_gemm_h100.cu
- benchmark.py
- Makefile
- ag_gemm_fp8_b200.cu
- benchmark.py
- Makefile
- all_gather.cu
- benchmark.py
- Makefile
- all_reduce.cu
- benchmark.py
- Makefile
- all_reduce_educational.cu
- Makefile
- all_to_all.cu
- benchmark.py
- Makefile
- benchmark.py
- gemm_ar_h100.cu
- gemm_ar_h100_lcsc.cu
- Makefile
- benchmark.py
- gemm_rs_b200.cu
- gemm_rs_h100.cu
- Makefile
- benchmark.py
- gemm_rs_fp8_b200.cu
- Makefile
- benchmark.py
- Makefile
- moe_dispatch_gemm_h100.cu
- benchmark.py
- Makefile
- reduce_scatter.cu
- benchmark.py
- Makefile
- ring_attn_h100.cu
- benchmark.py
- Makefile
- ulysses_attn.cu
- common.py
- README.md
- rotary.py
- triton_rotary.py
- benchmark.py
- gentests.py
- harness.impl
- harness2.impl
- Makefile
- rotary.cu
- test_correctness.py
- common.mk
- common.cuh
- templates.cuh
- util.cuh
- interpreter.cuh
- templates.cuh
- lcf.cuh
- templates.cuh
- lcsc.cuh
- lcsf.cuh
- templates.cuh
- prototype.cuh
- global_to_register.cu
- global_to_register.cuh
- global_to_shared.cu
- global_to_shared.cuh
- shared_to_register.cu
- shared_to_register.cuh
- tensor_to_register.cu
- tensor_to_register.cuh
- tile.cu
- tile.cuh
- global_to_register.cu
- global_to_register.cuh
- global_to_shared.cu
- global_to_shared.cuh
- shared_to_register.cu
- shared_to_register.cuh
- vec.cu
- vec.cuh
- memory.cu
- memory.cuh
- mma.cu
- mma.cuh
- tensor.cu
- tensor.cuh
- complex.cu
- complex.cuh
- mma.cu
- mma.cuh
- mma.cu
- mma.cuh
- warp.cu
- warp.cuh
- complex.cu
- complex.cuh
- fp16_fp16.cu
- fp16_fp16.cuh
- fp32_bf16.cu
- fp32_bf16.cuh
- fp32_fp16.cu
- fp32_fp16.cuh
- fp16_fp16.cu
- fp16_fp16.cuh
- fp16_fp8.cu
- fp16_fp8.cuh
- fp32_bf16.cu
- fp32_bf16.cuh
- fp32_fp16.cu
- fp32_fp16.cuh
- fp32_fp8.cu
- fp32_fp8.cuh
- int32_int8.cu
- int32_int8.cuh
- int32_uint8.cu
- int32_uint8.cuh
- warpgroup.cu
- warpgroup.cuh
- mma.cu
- mma.cuh
- complex_mul.cu
- complex_mul.cuh
- conversions.cu
- conversions.cuh
- maps.cu
- maps.cuh
- reductions.cu
- reductions.cuh
- tile.cu
- tile.cuh
- conversions.cu
- conversions.cuh
- maps.cu
- maps.cuh
- reductions.cu
- reductions.cuh
- vec.cu
- vec.cuh
- register.cu
- register.cuh
- conversions.cu
- conversions.cuh
- maps.cu
- maps.cuh
- reductions.cu
- reductions.cuh
- tile.cu
- tile.cuh
- conversions.cu
- conversions.cuh
- maps.cu
- maps.cuh
- reductions.cu
- reductions.cuh
- vec.cu
- vec.cuh
- shared.cu
- shared.cuh
- group.cu
- group.cuh
- testing_commons.cuh
- testing_flags.cuh
- testing_utils.cu
- testing_utils.cuh
- tile.cu
- tile.cuh
- tma.cu
- tma.cuh
- tma_multicast.cu
- tma_multicast.cuh
- tma_pgl.cu
- tma_pgl.cuh
- tma.cu
- tma.cuh
- tma_multicast.cu
- tma_multicast.cuh
- tma_pgl.cu
- tma_pgl.cuh
- vec.cu
- vec.cuh
- memory.cu
- memory.cuh
- thread.cu
- thread.cuh
- Makefile
- unit_tests.cu
- .gitignore
- Doxyfile
- LICENSE
- README.md
# 설치 가이드
1. 코드 내려받기
git clone https://github.com/HazyResearch/ThunderKittens
깃허브에서 프로젝트 코드 전체를 내 컴퓨터로 내려받습니다.
cd ThunderKittens
방금 내려받은 프로젝트 폴더 안으로 이동합니다.
2. 공식 설치 스크립트
쉬움 추천사전 준비물
- APT (Debian/Ubuntu 계열) Debian/Ubuntu 계열 리눅스에 기본 내장된 패키지 매니저입니다.
sudo apt install gcc-11 g++-11
APT 패키지 저장소에서 바로 설치합니다 (Debian/Ubuntu 계열).
sudo apt install clang-11
APT 패키지 저장소에서 바로 설치합니다 (Debian/Ubuntu 계열).
설치 후 새 터미널을 열고, 프로그램의 버전 확인 명령(예: --version)으로 정상 설치됐는지 확인하세요.
이 레포의 README에 적힌 실제 명령어를 그대로 가져왔습니다.
3. Make
보통사전 준비물
- Git GitHub에서 프로젝트 코드를 내려받으려면 필요합니다.
- Make Linux/macOS는 보통 기본 설치되어 있습니다. Windows는 별도 설치(예: MSYS2, WSL)가 필요합니다.
make
생성된 빌드 설정을 바탕으로 실제 컴파일을 진행해 실행 파일을 만듭니다.
make run
생성된 빌드 설정을 바탕으로 실제 컴파일을 진행해 실행 파일을 만듭니다.
에러 없이 끝나면 성공입니다. 생성된 실행 파일을 직접 실행해보세요.
이 레포의 README에 적힌 실제 명령어를 그대로 가져왔습니다.
// repository documentation
Was this content helpful?
(0 ratings)
