Skip to content

fix: Reduce memory usage in fused moe op associated with AutoTuning and fix moe fallback issue. #3793

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

Merged
merged 2 commits into from
Apr 24, 2025
Merged

fix: Reduce memory usage in fused moe op associated with AutoTuning and fix moe fallback issue. #3793

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
7822945
Reduce memory usage in fused moe op associated with AutoTuning.
hyukn Apr 23, 2025
b5abcf8
Fix fused_moe fallback issue. (#3652)
hyukn Apr 17, 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
34 changes: 18 additions & 16 deletions cpp/tensorrt_llm/thop/moeOp.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -161,17 +161,11 @@ class FusedMoeRunner : public torch::CustomClassHolder
torch::optional<torch::Tensor> input_sf, int64_t const tp_size, int64_t const tp_rank, int64_t const ep_size,
int64_t const ep_rank, bool min_latency_mode, torch::optional<c10::ArrayRef<int64_t>> profile_ids)
{
// Free the profile workspace to save memory
if (mProfileWorkspace != nullptr)
{
auto const cu_free_status = cudaFree(mProfileWorkspace);
TORCH_CHECK(
cu_free_status == cudaSuccess, "Can't free profile workspace for MoE GEMM profile before runMoe.");
mProfileWorkspace = nullptr;
}

std::lock_guard<std::mutex> lock(mMutex);

// Free the profile workspace to save memory
freeProfileWorkspace();

CHECK_INPUT(input, mActivationDtype)
CHECK_INPUT(token_selected_experts, at::ScalarType::Int)
if (token_final_scales)
Expand Down Expand Up @@ -248,6 +242,9 @@ class FusedMoeRunner : public torch::CustomClassHolder
{
std::lock_guard<std::mutex> lock(mMutex);

// Free the profile workspace to save memory
freeProfileWorkspace();

CHECK_INPUT(input, mActivationDtype)
CHECK_INPUT(token_selected_experts, at::ScalarType::Int)
if (token_final_scales)
Expand Down Expand Up @@ -375,13 +372,7 @@ class FusedMoeRunner : public torch::CustomClassHolder
hidden_size, inter_size, GROUP_SIZE, tensorrt_llm::ActivationType::Swiglu, USE_BIAS, USE_LORA,
min_latency_mode, parallelism_config);

if (mProfileWorkspace != nullptr)
{
auto const cu_free_status = cudaFree(mProfileWorkspace);
TORCH_CHECK(cu_free_status == cudaSuccess,
"Can't free profile workspace for MoE GEMM profile during memory reallocation.");
mProfileWorkspace = nullptr;
}
freeProfileWorkspace();
size_t profile_workspace_size = mProfiler->getWorkspaceSize(num_rows);
auto const cu_malloc_status = cudaMalloc(&mProfileWorkspace, profile_workspace_size);
TORCH_CHECK(cu_malloc_status == cudaSuccess, "Can't allocate profile workspace for MoE GEMM profile.");
Expand Down Expand Up @@ -416,6 +407,17 @@ class FusedMoeRunner : public torch::CustomClassHolder
using Profile = tensorrt_llm::cutlass_extensions::CutlassGemmConfig;
std::vector<Profile> mAllProfiles;

void freeProfileWorkspace()
{
if (mProfileWorkspace != nullptr)
{
auto const cu_free_status = cudaFree(mProfileWorkspace);
TORCH_CHECK(cu_free_status == cudaSuccess,
"Can't free profile workspace for MoE GEMM profile during memory reallocation.");
mProfileWorkspace = nullptr;
}
}

void setRunnerProfiles(torch::optional<c10::ArrayRef<int64_t>> profile_ids)
{
if (mUseFp8BlockScaling)
Expand Down
8 changes: 5 additions & 3 deletions tensorrt_llm/_torch/custom_ops/torch_custom_ops.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -120,13 +120,15 @@ def fused_moe(
# TODO: only profile for min_latency_mode = False due to the error in the moe_kernels
tuning_config = TuningConfig(dynamic_tensors=(
# input, dim 0, all valid buckets, map a seq_len to power of 2 bucket index
(0, 0, ((16384, 8192, 4096, 2048, 1024, 512, 256, 128, 64, 32, 16, 8, 4,
2, 1), next_positive_power_of_2)),
(0, 0, ((8192, 4096, 2048, 1024, 512, 256, 128, 64, 32, 16, 8, 4, 2, 1),
next_positive_power_of_2)),
# min_latency_tensor, dim 0, (0 for False, 1 for True), map to it self
(2, 0, ((0, ), lambda x: x)),
))

min_latency_tensor = torch.empty(1) if min_latency_mode else torch.empty(0)
# TODO: set min_latency_mode always to False due to the error in the moe_kernels
min_latency_tensor = torch.empty(0)

# allocate workspace for profiling
moe_runner = MoERunner(
x_dtype=input.dtype,
Expand Down
2 changes: 1 addition & 1 deletion tensorrt_llm/_torch/utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -197,7 +197,7 @@ def get_power_of_2_num_tokens_buckets(max_num_tokens) -> List[int]:
num_token_buckets.append(m)
m //= 2

return num_token_buckets
return tuple(num_token_buckets)


def get_last_power_of_2_num_tokens_buckets(max_num_tokens) -> List[int]:
Expand Down