Skip to content

Safe VLM JSON config parsing with read_json_param() #2785

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 3 commits into from
Oct 6, 2025
Merged

Safe VLM JSON config parsing with read_json_param() #2785

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
5007ea2
Extend read_json_param to handle small arrays, add std array type check
yatarkan Oct 2, 2025
286097f
Use read_json_param for safe processor config parsing
yatarkan Oct 2, 2025
c3beb25
Use read_json_param for safe vlm config parsing
yatarkan 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
10 changes: 9 additions & 1 deletion src/cpp/src/json_utils.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -14,12 +14,20 @@ namespace ov {
namespace genai {
namespace utils {

template<typename, typename = void>
constexpr bool is_std_array = false;

template<typename T, std::size_t N>
constexpr bool is_std_array<std::array<T, N>> = true;

/// @brief reads value to param if T argument type is aligned with value stores in json
/// if types are not compatible leave param unchanged
template <typename T>
void read_json_param(const nlohmann::json& data, const std::string& name, T& param) {
if (data.contains(name)) {
if (data[name].is_number() || data[name].is_boolean() || data[name].is_string() || data[name].is_object()) {
if (data[name].is_number() || data[name].is_boolean() || data[name].is_string() || data[name].is_object()
|| (is_std_array<T> && data[name].is_array())
Comment on lines +28 to +29
Copy link

Copilot AI Oct 2, 2025

Choose a reason for hiding this comment

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

[nitpick] The condition logic could be simplified and made more readable by extracting the array condition into a separate variable or reorganizing the boolean expressions for better clarity.

Suggested change
if (data[name].is_number() || data[name].is_boolean() || data[name].is_string() || data[name].is_object()
|| (is_std_array<T> && data[name].is_array())
bool is_compatible_array = is_std_array<T> && data[name].is_array();
if (data[name].is_number() || data[name].is_boolean() || data[name].is_string() || data[name].is_object()
|| is_compatible_array

Copilot uses AI. Check for mistakes.

) {
param = data[name].get<T>();
}
} else if (name.find(".") != std::string::npos) {
Expand Down
46 changes: 14 additions & 32 deletions src/cpp/src/visual_language/processor_config.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -14,37 +14,23 @@ ov::genai::ProcessorConfig::ProcessorConfig(const std::filesystem::path& json_pa
read_json_param(parsed, "patch_size", patch_size); // For llava - stored in config.json vision_config
read_json_param(parsed, "scale_resolution", scale_resolution);
read_json_param(parsed, "max_slice_nums", max_slice_nums);
if (parsed.contains("norm_mean")) {
norm_mean = parsed.at("norm_mean").get<std::array<float, 3>>();
}
if (parsed.contains("norm_std")) {
norm_std = parsed.at("norm_std").get<std::array<float, 3>>();
}

read_json_param(parsed, "norm_mean", norm_mean);
read_json_param(parsed, "norm_std", norm_std);

// Setting llava config params
if (parsed.contains("image_mean")) {
image_mean = parsed.at("image_mean").get<std::array<float, 3>>();
}
if (parsed.contains("image_std")) {
image_std = parsed.at("image_std").get<std::array<float, 3>>();
}

if (parsed.contains("crop_size")) {
crop_size_height = parsed.at("crop_size").at("height");
crop_size_width = parsed.at("crop_size").at("width");
}
if (parsed.contains("size") && parsed.at("size").contains("shortest_edge")) {
size_shortest_edge = parsed.at("size").at("shortest_edge");
}
read_json_param(parsed, "image_mean", image_mean);
read_json_param(parsed, "image_std", image_std);
read_json_param(parsed, "crop_size.height", crop_size_height);
read_json_param(parsed, "crop_size.width", crop_size_width);
read_json_param(parsed, "size.shortest_edge", size_shortest_edge);

// Setting llava-next config params
if (parsed.contains("image_grid_pinpoints")) {
image_grid_pinpoints = parsed.at("image_grid_pinpoints").get<std::vector<std::pair<int, int>>>();
}
read_json_param(parsed, "image_grid_pinpoints", image_grid_pinpoints);
read_json_param(parsed, "num_crops", phi3_v.num_crops);
if (parsed.contains("img_processor")) {
phi3_v.num_img_tokens = parsed.at("img_processor").at("num_img_tokens");
}

// Setting phi3_v config params
read_json_param(parsed, "img_processor.num_img_tokens", phi3_v.num_img_tokens);

// Setting qwen2vl config params
read_json_param(parsed, "min_pixels", min_pixels);
Expand All @@ -53,10 +39,6 @@ ov::genai::ProcessorConfig::ProcessorConfig(const std::filesystem::path& json_pa
read_json_param(parsed, "merge_size", merge_size);

// Setting gemma3-4b-it config params
if (parsed.contains("size") && parsed.at("size").contains("height")) {
size_height = parsed.at("size").at("height");
}
if (parsed.contains("size") && parsed.at("size").contains("width")) {
size_width = parsed.at("size").at("width");
}
read_json_param(parsed, "size.height", size_height);
read_json_param(parsed, "size.width", size_width);
}
14 changes: 6 additions & 8 deletions src/cpp/src/visual_language/vlm_config.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -53,22 +53,20 @@ VLMConfig::VLMConfig(const std::filesystem::path& json_path) {

// Setting llava_next specific config params
read_json_param(parsed, "image_newline", image_newline);
if (parsed.contains("vision_config")) {
read_json_param(parsed.at("vision_config"), "patch_size", vision_config_patch_size);
}
read_json_param(parsed, "vision_config.patch_size", vision_config_patch_size);

// phi3_v and phi4mm
if (parsed.contains("sub_GN")) {
if (parsed.contains("sub_GN") && parsed.at("sub_GN").is_array()) {
sub_GN = parsed.at("sub_GN").get<std::vector<std::vector<std::vector<std::vector<float>>>>>().at(0).at(0).at(0);
}
assert_size(sub_GN.size(), model_type);
if (parsed.contains("glb_GN")) {
if (parsed.contains("glb_GN") && parsed.at("glb_GN").is_array()) {
glb_GN = parsed.at("glb_GN").get<std::vector<std::vector<std::vector<float>>>>().at(0).at(0);
}
assert_size(glb_GN.size(), model_type);

// Qwen2.5VL
if (parsed.contains("vision_config")) {
read_json_param(parsed.at("vision_config"), "window_size", vision_config_window_size);
}
read_json_param(parsed, "vision_config.window_size", vision_config_window_size);
}

} // namespace ov::genai
Loading