Skip to content

bugfix: add calculation of mm_embedding size in shm manager. #1298

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
shan-chen-feng wants to merge 2 commits into
base: main
Choose a base branch
from
+27 −31
Open

bugfix: add calculation of mm_embedding size in shm manager. #1298

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
0cbad1c
bugfix: add calculation of mm_embedding size in shm manager.
shan-chen-feng Apr 16, 2026
a275173
perf: remove additional data transfer for mm_embedding & dit inputs.
shan-chen-feng Apr 17, 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
9 changes: 5 additions & 4 deletions xllm/core/runtime/embed_vlm_worker_impl.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -84,7 +84,6 @@ std::optional<ForwardOutput> EmbedVLMWorkerImpl::step(
input.sampling_params.is_embeddings) {
auto embeddings =
model_->pooler(hidden_states, sampling_params.selected_token_idxes);
sample_output.embeddings = embeddings;
// split full embeddings and add them to mm_embeddings
// so that the user could receive embeddings of images and texts
if (FLAGS_enable_return_mm_full_embeddings) {
Expand All @@ -97,10 +96,12 @@ std::optional<ForwardOutput> EmbedVLMWorkerImpl::step(
sample_output.mm_embeddings.emplace_back(image_embed);
token_start_idx += seq_len;
}
output.sample_output = sample_output;
} else {
sample_output.embeddings = embeddings;
output.sample_output = sample_output;
output.embedding = embeddings;
Copy link
Copy Markdown
Collaborator

@wly-115 wly-115 Apr 17, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Can we remove the output.embedding = embeddings ？

}

output.sample_output = sample_output;
output.embedding = embeddings;
}
ret = device_.synchronize_default_stream();
return output;
Expand Down
49 changes: 22 additions & 27 deletions xllm/core/runtime/forward_shared_memory_manager.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -344,9 +344,10 @@ inline size_t get_dit_forward_input_size(const DiTForwardInput& input) {
return size;
}

inline size_t get_dit_forward_output_size(const DiTForwardOutput& output) {
size_t size = type_size<uint64_t>; // vector size
for (const auto& tensor : output.tensors) {
inline size_t get_vector_tensor_size(
Copy link
Copy Markdown
Contributor

@gemini-code-assist gemini-code-assist Bot Apr 16, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

high

According to the repository style guide (Rule 177), file-local functions used only within a single .cpp file must be placed in an anonymous namespace to ensure internal linkage and avoid potential symbol collisions. Since get_vector_tensor_size is not declared in the header file, it should be moved into an anonymous namespace.

References
  1. File-local functions and variables (used only within a single .cpp file) must be placed in an anonymous namespace. (link)

const std::vector<torch::Tensor>& tensor_vec) {
size_t size = type_size<int32_t>; // vector size
for (const auto& tensor : tensor_vec) {
size += get_tensor_size(tensor);
}
return size;
Expand Down Expand Up @@ -1017,14 +1018,6 @@ inline void write_dit_forward_input(RawInputSerializeContext& context,
write_dit_generation_params(context, input.generation_params);
}

inline void write_dit_forward_output(char*& buffer,
const DiTForwardOutput& output) {
write_data(buffer, static_cast<uint64_t>(output.tensors.size()));
for (const auto& tensor : output.tensors) {
write_tensor(buffer, tensor);
}
}

inline void safe_advance_buffer(const char*& buffer, size_t offset) {
if (buffer != nullptr) {
buffer += offset;
Expand Down Expand Up @@ -1876,16 +1869,6 @@ inline void read_dit_forward_input(ReadContext& context,
read_dit_generation_params(context, input.generation_params);
}

inline void read_dit_forward_output(const char*& buffer,
DiTForwardOutput& output) {
uint64_t size;
read_data(buffer, size);
output.tensors.resize(size);
for (auto& tensor : output.tensors) {
read_tensor(buffer, tensor);
}
}

inline void initialize_device_buffer_session(ReadContext& context,
ForwardInput& forward_input,
const torch::Device& device,
Expand Down Expand Up @@ -2094,7 +2077,9 @@ inline void deserialize_raw_forward_input(const char*& buffer,
/*force_host_materialize=*/true);
#endif

read_dit_forward_input(context, input_params.dit_forward_input);
if (FLAGS_backend == "dit") {
read_dit_forward_input(context, input_params.dit_forward_input);
}

finalize_device_buffer_session(device_session, stream);
buffer = context.tensor_cursor;
Expand Down Expand Up @@ -2177,8 +2162,9 @@ inline void serialize_raw_forward_input_sections(

write_vector_to_tensor(context, input.new_token_slot_ids);
write_2d_vector_to_tensor(context, input.block_tables_vec);

write_dit_forward_input(context, input.dit_forward_input);
if (FLAGS_backend == "dit") {
write_dit_forward_input(context, input.dit_forward_input);
}
}

inline RawInputLayoutHeader calculate_raw_forward_input_layout(
Expand Down Expand Up @@ -2260,8 +2246,12 @@ size_t calculate_raw_forward_output_size(const RawForwardOutput& output) {
size += get_vector_size(output.out_tokens);
size += get_vector_size(output.out_logprobs);
size += type_size<int32_t>; // prepared_layer_id
// mm_embedding_data
size += get_vector_tensor_size(output.mm_embeddings);
// dit output data
size += get_dit_forward_output_size(output.dit_forward_output);
if (FLAGS_backend == "dit") {
size += get_vector_tensor_size(output.dit_forward_output.tensors);
}

return size;
}
Expand Down Expand Up @@ -2327,8 +2317,11 @@ void deserialize_raw_forward_output(const char* buffer,
read_data(buffer, output.prepared_layer_id);

read_vector_tensor(buffer, output.mm_embeddings);

// read dit output
read_dit_forward_output(buffer, output.dit_forward_output);
if (FLAGS_backend == "dit") {
read_vector_tensor(buffer, output.dit_forward_output.tensors);
}
}

void serialize_raw_forward_output(const RawForwardOutput& output,
Expand All @@ -2344,7 +2337,9 @@ void serialize_raw_forward_output(const RawForwardOutput& output,

write_vector_tensor(buffer, output.mm_embeddings);
// write dit output
write_dit_forward_output(buffer, output.dit_forward_output);
if (FLAGS_backend == "dit") {
write_vector_tensor(buffer, output.dit_forward_output.tensors);
}
}

void convert_raw_forward_input_to_forward_input(RawForwardInput& raw_input,
Expand Down
Loading