Skip to content

Update the routing for TRTLLMGEN to support kimi k2 and qwen #1831

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 4 commits into
base: main
Choose a base branch
from
+964 −504
Open

Update the routing for TRTLLMGEN to support kimi k2 and qwen #1831

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
e8d7e2b
Update the routing for TRTLLMGEN to support kimi k2 and qwen
ChristinaZ Oct 1, 2025
49297c7
fix compile error
jiahanc Oct 1, 2025
1adc570
fix format
jiahanc Oct 2, 2025
c9e42ca
update some change
jiahanc Oct 2, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
453 changes: 272 additions & 181 deletions csrc/trtllm_fused_moe_routing_deepseek.cu
View file
Open in desktop

Large diffs are not rendered by default.

227 changes: 143 additions & 84 deletions csrc/trtllm_fused_moe_routing_llama4.cu
View file
Open in desktop

Large diffs are not rendered by default.

329 changes: 251 additions & 78 deletions csrc/trtllm_fused_moe_routing_renormalize.cu
View file
Open in desktop

Large diffs are not rendered by default.

84 changes: 53 additions & 31 deletions include/flashinfer/trtllm/fused_moe/DevKernel.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -112,39 +112,61 @@ namespace moe::dev {
FLASHINFER_WARN("Unsupported pair"); \
}

#define LAUNCH_ROUTING(data, coopLaunch, kernel, numBlocks, numThreads, smemSize, stream) \
if (data.mDtypeExpW == tg::Dtype::Fp32) { \
LAUNCH_PDL(data, coopLaunch, LAUNCH_ESC(float, float), kernel, numBlocks, numThreads, \
smemSize, stream); \
} else if (data.mDtypeExpW == tg::Dtype::Bfloat16) { \
LAUNCH_PDL(data, coopLaunch, LAUNCH_ESC(__nv_bfloat16, __nv_bfloat16), kernel, numBlocks, \
numThreads, smemSize, stream); \
} else { \
FLASHINFER_WARN("Unsupported dtypeExpW"); \
#define LAUNCH_ROUTING_LLAMA4(data, coopLaunch, kernel, numBlocks, numThreads, smemSize, stream) \
if (data.mDtypeExpW == tg::Dtype::Fp32) { \
LAUNCH_PDL(data, coopLaunch, LAUNCH_ESC(float, float, 128 /* Always 128 for llama4*/), kernel, \
numBlocks, numThreads, smemSize, stream); \
} else if (data.mDtypeExpW == tg::Dtype::Bfloat16) { \
LAUNCH_PDL(data, coopLaunch, \
LAUNCH_ESC(__nv_bfloat16, __nv_bfloat16, 128 /* Always 128 for llama4*/), kernel, \
numBlocks, numThreads, smemSize, stream); \
} else { \
TLLM_LOG_ERROR("Unsupported dtypeExpW"); \
}

#define LAUNCH_ROUTING_WITH_EXTRA_FLAG(data, coopLaunch, kernel, numBlocks, numThreads, smemSize, \
stream, extraFlag, forceFloatInput) \
if (data.mDtypeExpW == tg::Dtype::Fp32 && extraFlag) { \
LAUNCH_PDL(data, coopLaunch, LAUNCH_ESC(float, float, true), kernel, numBlocks, numThreads, \
smemSize, stream); \
} else if (data.mDtypeExpW == tg::Dtype::Fp32) { \
LAUNCH_PDL(data, coopLaunch, LAUNCH_ESC(float, float, false), kernel, numBlocks, numThreads, \
smemSize, stream); \
} else if (data.mDtypeExpW == tg::Dtype::Bfloat16 && extraFlag && forceFloatInput) { \
LAUNCH_PDL(data, coopLaunch, LAUNCH_ESC(float, __nv_bfloat16, true), kernel, numBlocks, \
numThreads, smemSize, stream); \
} else if (data.mDtypeExpW == tg::Dtype::Bfloat16 && extraFlag) { \
LAUNCH_PDL(data, coopLaunch, LAUNCH_ESC(__nv_bfloat16, __nv_bfloat16, true), kernel, \
numBlocks, numThreads, smemSize, stream); \
} else if (data.mDtypeExpW == tg::Dtype::Bfloat16 && forceFloatInput) { \
LAUNCH_PDL(data, coopLaunch, LAUNCH_ESC(float, __nv_bfloat16, false), kernel, numBlocks, \
numThreads, smemSize, stream); \
} else if (data.mDtypeExpW == tg::Dtype::Bfloat16) { \
LAUNCH_PDL(data, coopLaunch, LAUNCH_ESC(__nv_bfloat16, __nv_bfloat16, false), kernel, \
numBlocks, numThreads, smemSize, stream); \
} else { \
FLASHINFER_WARN("Unsupported dtypeExpW"); \
#define LAUNCH_ROUTING_WITH_NUM_EXPERTS_FORCE_FLOAT_INPUT(data, coopLaunch, kernel, numBlocks, \
numThreads, smemSize, stream, extraFlag, \
forceFloatInput, numExperts) \
if (data.mDtypeExpW == tg::Dtype::Fp32 && extraFlag) { \
LAUNCH_PDL(data, coopLaunch, LAUNCH_ESC(float, float, numExperts, true), kernel, numBlocks, \
numThreads, smemSize, stream); \
} else if (data.mDtypeExpW == tg::Dtype::Fp32) { \
LAUNCH_PDL(data, coopLaunch, LAUNCH_ESC(float, float, numExperts, false), kernel, numBlocks, \
numThreads, smemSize, stream); \
} else if (data.mDtypeExpW == tg::Dtype::Bfloat16 && extraFlag && forceFloatInput) { \
LAUNCH_PDL(data, coopLaunch, LAUNCH_ESC(float, __nv_bfloat16, numExperts, true), kernel, \
numBlocks, numThreads, smemSize, stream); \
} else if (data.mDtypeExpW == tg::Dtype::Bfloat16 && extraFlag) { \
LAUNCH_PDL(data, coopLaunch, LAUNCH_ESC(__nv_bfloat16, __nv_bfloat16, numExperts, true), \
kernel, numBlocks, numThreads, smemSize, stream); \
} else if (data.mDtypeExpW == tg::Dtype::Bfloat16 && forceFloatInput) { \
LAUNCH_PDL(data, coopLaunch, LAUNCH_ESC(float, __nv_bfloat16, numExperts, false), kernel, \
numBlocks, numThreads, smemSize, stream); \
} else if (data.mDtypeExpW == tg::Dtype::Bfloat16) { \
LAUNCH_PDL(data, coopLaunch, LAUNCH_ESC(__nv_bfloat16, __nv_bfloat16, numExperts, false), \
kernel, numBlocks, numThreads, smemSize, stream); \
} else { \
TLLM_LOG_ERROR("Unsupported dtypeExpW"); \
}

////////////////////////////////////////////////////////////////////////////////////////////////////

#define LAUNCH_ROUTING_WITH_NUM_EXPERTS(data, coopLaunch, kernel, numBlocks, numThreads, smemSize, \
stream, extraFlag1, numExperts) \
if (data.mDtypeExpW == tg::Dtype::Fp32 && extraFlag1) { \
LAUNCH_PDL(data, coopLaunch, LAUNCH_ESC(float, float, numExperts, true), kernel, numBlocks, \
numThreads, smemSize, stream); \
} else if (data.mDtypeExpW == tg::Dtype::Fp32) { \
LAUNCH_PDL(data, coopLaunch, LAUNCH_ESC(float, float, numExperts, false), kernel, numBlocks, \
numThreads, smemSize, stream); \
} else if (data.mDtypeExpW == tg::Dtype::Bfloat16 && extraFlag1) { \
LAUNCH_PDL(data, coopLaunch, LAUNCH_ESC(__nv_bfloat16, __nv_bfloat16, numExperts, true), \
kernel, numBlocks, numThreads, smemSize, stream); \
} else if (data.mDtypeExpW == tg::Dtype::Bfloat16) { \
LAUNCH_PDL(data, coopLaunch, LAUNCH_ESC(__nv_bfloat16, __nv_bfloat16, numExperts, false), \
kernel, numBlocks, numThreads, smemSize, stream); \
} else { \
TLLM_LOG_ERROR("Unsupported dtypeExpW"); \
}

////////////////////////////////////////////////////////////////////////////////////////////////////
Expand Down
Loading