Skip to content

fix: use type-specific FP8 max value for clamping in RMSNorm quantization kernels #2612

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
Bias92 wants to merge 2 commits into
base: main
Choose a base branch
from
+19 −2
Open

fix: use type-specific FP8 max value for clamping in RMSNorm quantization kernels #2612

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
11088ee
fix: use type-specific FP8 max value for clamping in RMSNorm quantiza…
Bias92 Feb 21, 2026
a45c497
style: fix pre-commit issues
Bias92 Feb 23, 2026
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
21 changes: 19 additions & 2 deletions include/flashinfer/norm.cuh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -145,6 +145,21 @@ cudaError_t RMSNorm(T* input, T* weight, T* output, uint32_t batch_size, uint32_
return cudaSuccess;
}

template <typename T>
struct fp8_clamp_max {
static_assert(sizeof(T) == 0, "Unsupported FP8 type for fp8_clamp_max");
};

template <>
struct fp8_clamp_max<__nv_fp8_e4m3> {
static constexpr float value = 448.0f;
};

template <>
struct fp8_clamp_max<__nv_fp8_e5m2> {
static constexpr float value = 57344.0f;
};

template <uint32_t VEC_SIZE, typename T, typename O>
__global__ void RMSNormQuantKernel(T* __restrict__ input, T* __restrict__ weight,
O* __restrict__ output, const uint32_t d,
Expand Down Expand Up @@ -214,7 +229,8 @@ __global__ void RMSNormQuantKernel(T* __restrict__ input, T* __restrict__ weight
for (uint32_t j = 0; j < VEC_SIZE; j++) {
output_vec[j] =
float(input_vec[j]) * rms_rcp * (weight_bias + float(weight_vec[j])) * scale_inv;
output_vec[j] = fmaxf(-448.0f, fminf(output_vec[j], 448.0f));
output_vec[j] =
fmaxf(-fp8_clamp_max<O>::value, fminf(output_vec[j], fp8_clamp_max<O>::value));
}
if ((i * num_threads + thread_id) * VEC_SIZE < d) {
output_vec.cast_store(output + bx * stride_output + i * num_threads * VEC_SIZE +
Expand Down Expand Up @@ -598,7 +614,8 @@ __global__ void FusedAddRMSNormQuantKernel(T* __restrict__ input, T* __restrict_
#pragma unroll
for (uint32_t j = 0; j < VEC_SIZE; j++) {
output_vec[j] = x_vec[j] * rms_rcp * (weight_bias + float(weight_vec[j])) * scale_inv;
output_vec[j] = fmaxf(-448.0f, fminf(output_vec[j], 448.0f));
output_vec[j] =
fmaxf(-fp8_clamp_max<O>::value, fminf(output_vec[j], fp8_clamp_max<O>::value));
}
if ((i * num_threads + thread_id) * VEC_SIZE < d) {
output_vec.cast_store(output + bx * stride_output + i * num_threads * VEC_SIZE +
Expand Down
Loading