messages listlengths 3 3 | task stringlengths 35 39 | source stringclasses 1
value | lang stringclasses 2
values | speedup float64 0.02 5.42 |
|---|---|---|---|---|
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/cuda_agent_ops_6k/01219.py | rft_r1 | Triton | 5.415498 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/cuda_agent_ops_6k/01219.py | rft_r1 | Triton | 4.967145 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/drkernel_rl/cuda_llm_617176.py | rft_r1 | Triton | 4.597948 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/drkernel_rl/cuda_llm_929891.py | rft_r1 | Triton | 3.472791 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/drkernel_rl/cuda_llm_929891.py | rft_r1 | Triton | 3.423805 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/drkernel_rl/cuda_llm_812302.py | rft_r1 | CUDA | 2.691519 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/cuda_agent_ops_6k/01239.py | rft_r1 | CUDA | 2.220569 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/drkernel_rl/cuda_llm_342840.py | rft_r1 | Triton | 1.778672 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/drkernel_rl/cuda_llm_342840.py | rft_r1 | CUDA | 1.772704 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/drkernel_rl/cuda_llm_251426.py | rft_r1 | CUDA | 1.474552 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/drkernel_rl/cuda_llm_270744.py | rft_r1 | CUDA | 1.329916 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/drkernel_rl/cuda_llm_334747.py | rft_r1 | CUDA | 1.268106 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/drkernel_rl/cuda_llm_841392.py | rft_r1 | Triton | 1.246056 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/drkernel_rl/cuda_llm_679297.py | rft_r1 | CUDA | 1.099259 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/drkernel_rl/cuda_llm_966419.py | rft_r1 | Triton | 1.072808 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/drkernel_rl/cuda_llm_188803.py | rft_r1 | Triton | 1.062701 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/drkernel_rl/cuda_llm_226644.py | rft_r1 | CUDA | 1.047683 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/drkernel_rl/cuda_llm_416708.py | rft_r1 | Triton | 1.015635 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/drkernel_rl/cuda_llm_137514.py | rft_r1 | Triton | 1.011439 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/drkernel_rl/cuda_llm_416708.py | rft_r1 | CUDA | 1.007899 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/drkernel_rl/cuda_llm_841392.py | rft_r1 | Triton | 1.007117 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/cuda_agent_ops_6k/00735.py | rft_r1 | Triton | 1.004228 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/drkernel_rl/cuda_llm_228671.py | rft_r1 | Triton | 1.004219 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/cuda_agent_ops_6k/00735.py | rft_r1 | Triton | 1.003929 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/cuda_agent_ops_6k/00296.py | rft_r1 | CUDA | 1.001342 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/drkernel_rl/cuda_llm_25017.py | rft_r1 | CUDA | 0.999896 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/cuda_agent_ops_6k/01439.py | rft_r1 | CUDA | 0.997872 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/drkernel_rl/cuda_llm_158099.py | rft_r1 | Triton | 0.994937 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/cuda_agent_ops_6k/01439.py | rft_r1 | CUDA | 0.991592 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/drkernel_rl/cuda_llm_158099.py | rft_r1 | Triton | 0.991084 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/cuda_agent_ops_6k/01439.py | rft_r1 | Triton | 0.990699 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/cuda_agent_ops_6k/01807.py | rft_r1 | Triton | 0.990589 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/cuda_agent_ops_6k/00465.py | rft_r1 | Triton | 0.986629 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/cuda_agent_ops_6k/00230.py | rft_r1 | CUDA | 0.983022 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/cuda_agent_ops_6k/00045.py | rft_r1 | Triton | 0.973684 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/cuda_agent_ops_6k/00045.py | rft_r1 | Triton | 0.97348 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/cuda_agent_ops_6k/00475.py | rft_r1 | Triton | 0.973011 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/cuda_agent_ops_6k/00045.py | rft_r1 | CUDA | 0.97178 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/cuda_agent_ops_6k/03117.py | rft_r1 | Triton | 0.958057 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/cuda_agent_ops_6k/02535.py | rft_r1 | CUDA | 0.956106 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/cuda_agent_ops_6k/01659.py | rft_r1 | Triton | 0.950731 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/cuda_agent_ops_6k/01361.py | rft_r1 | CUDA | 0.939761 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/drkernel_rl/cuda_llm_226644.py | rft_r1 | CUDA | 0.931288 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/cuda_agent_ops_6k/01726.py | rft_r1 | CUDA | 0.929403 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/drkernel_rl/cuda_llm_83069.py | rft_r1 | Triton | 0.926656 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/drkernel_rl/cuda_llm_841392.py | rft_r1 | Triton | 0.923638 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/cuda_agent_ops_6k/02436.py | rft_r1 | CUDA | 0.909774 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/drkernel_rl/cuda_llm_342840.py | rft_r1 | CUDA | 0.882936 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/drkernel_rl/cuda_llm_931543.py | rft_r1 | Triton | 0.863378 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/drkernel_rl/cuda_llm_310045.py | rft_r1 | CUDA | 0.855331 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/drkernel_rl/cuda_llm_163577.py | rft_r1 | CUDA | 0.852184 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/drkernel_rl/cuda_llm_416708.py | rft_r1 | Triton | 0.842013 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/drkernel_rl/cuda_llm_182390.py | rft_r1 | Triton | 0.832181 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/drkernel_rl/cuda_llm_966419.py | rft_r1 | CUDA | 0.754976 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/drkernel_rl/cuda_llm_334747.py | rft_r1 | CUDA | 0.640924 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/cuda_agent_ops_6k/00738.py | rft_r1 | Triton | 0.635678 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/cuda_agent_ops_6k/00546.py | rft_r1 | CUDA | 0.581691 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/cuda_agent_ops_6k/03117.py | rft_r1 | CUDA | 0.526226 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/cuda_agent_ops_6k/00204.py | rft_r1 | CUDA | 0.482697 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/cuda_agent_ops_6k/00733.py | rft_r1 | CUDA | 0.478612 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/drkernel_rl/cuda_llm_958825.py | rft_r1 | CUDA | 0.456035 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/drkernel_rl/cuda_llm_563306.py | rft_r1 | CUDA | 0.440847 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/drkernel_rl/cuda_llm_188803.py | rft_r1 | CUDA | 0.407572 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/cuda_agent_ops_6k/01159.py | rft_r1 | CUDA | 0.386361 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/drkernel_rl/cuda_llm_391023.py | rft_r1 | CUDA | 0.352465 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/cuda_agent_ops_6k/00475.py | rft_r1 | CUDA | 0.331126 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/drkernel_rl/cuda_llm_265224.py | rft_r1 | CUDA | 0.309153 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/cuda_agent_ops_6k/00211.py | rft_r1 | Triton | 0.304516 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/cuda_agent_ops_6k/00735.py | rft_r1 | Triton | 0.301335 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/drkernel_rl/cuda_llm_265224.py | rft_r1 | CUDA | 0.275271 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/cuda_agent_ops_6k/02535.py | rft_r1 | Triton | 0.237273 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/drkernel_rl/cuda_llm_954860.py | rft_r1 | CUDA | 0.232422 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/drkernel_rl/cuda_llm_747164.py | rft_r1 | CUDA | 0.152708 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/cuda_agent_ops_6k/00738.py | rft_r1 | Triton | 0.141244 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/cuda_agent_ops_6k/00614.py | rft_r1 | CUDA | 0.126943 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/drkernel_rl/cuda_llm_747164.py | rft_r1 | Triton | 0.103076 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/drkernel_rl/cuda_llm_993283.py | rft_r1 | CUDA | 0.066338 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/cuda_agent_ops_6k/01807.py | rft_r1 | Triton | 0.048566 |
[
{
"role": "system",
"content": "You are an expert GPU kernel engineer working on one NVIDIA H200 SXM (Hopper, sm_90).\nAvailable: torch, triton, flashinfer, deepgemm, tilelang, cute dsl, cutlass; native CUDA via torch.utils.cpp_extension.load_inline. No network.\nCode must be CUDA-graph capturable. Think ef... | external/drkernel_rl/cuda_llm_495241.py | rft_r1 | CUDA | 0.020712 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.