[WIP] [GSOC] GGUF Importer #27177
[WIP] [GSOC] GGUF Importer #27177
Conversation
| GGUFBuffer::GGUFBuffer(const std::string & fileName){ | ||
| std::ifstream file(fileName, std::ios::binary | std::ios::ate); | ||
| if (!file.is_open()) { | ||
| throw std::runtime_error("Could not open file: "); |
There was a problem hiding this comment.
Please use CV_Error, CV_Assert, etc. See https://docs.opencv.org/5.x/db/de0/group__core__utils.html#ga5b48c333c777666e076bd7052799f891
| if (!file.read(reinterpret_cast<char*>(buf.data()), size)) { | ||
| throw std::runtime_error("Error reading file: " ); | ||
| } |
There was a problem hiding this comment.
The same for error handling.
| GTEST_API_ int main(int argc, char **argv) | ||
| { | ||
| testing::InitGoogleTest(&argc, argv); | ||
| return RUN_ALL_TESTS(); | ||
| } |
There was a problem hiding this comment.
it should not be here. We have genetic test main to handle extra options and environment.
| { | ||
| // Locate the GGUF file; this should be in the directory dnn/gguf/ (adjust the filename as needed) | ||
| std::string ggufModelPath = _tf("mha.gguf", true); | ||
| std::string onnxModelPath = _tf("mha.onnx", true); |
There was a problem hiding this comment.
It makes sense to save the model output as .pb file with ONNX runtime for example, but not relay on OpenCV implementation for ONNX. It's more reliable and easy to catch regressions.
| std::vector<int> dims; | ||
| for (uint32_t i = 0; i < dim_count; ++i) { | ||
| dims.push_back(reader.readSingleValueInt<int64_t>()); | ||
| } |
There was a problem hiding this comment.
It's more efficient to use std::vector dims(dim_count); and assign values, rather than call push_back and trigger reallocations.
| if (tensor.type != GGML_TYPE_F32) { | ||
| throw std::runtime_error("Unsupported tensor type: " + std::to_string(tensor.type)); | ||
| } |
There was a problem hiding this comment.
Use CV_Assert and CV_Check
| tensor.role = get_tensor_role(layerName); | ||
| tensor.n_block = blockn; | ||
| } else { | ||
| throw std::runtime_error("Invalid tensor name format: " + tensor_name); |
| if (t.size() == 0) { | ||
| throw std::runtime_error("No input tensors found"); |
| if (t.size() == 0) { | ||
| throw std::runtime_error("No input tensors found"); | ||
| } |
| throw std::runtime_error( | ||
| "Unsupported tensor dimension: " + std::to_string(dims.size())); |
draft PR for #27176
Pull Request Readiness Checklist
See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request
Patch to opencv_extra has the same branch name.