Example Yolo weights conversion issue

[scamianbas]

After having downloaded weights from https://pjreddie.com/media/files/yolov3-tiny.weights (notice that sha1sum differs: ec08d0d75bfc4ef526291dc737b0b9242e479cf6)

Weights conversion is completed but with a runtime warning:

(.venv) root@bb47dd5d8081:/ggml/examples/yolo# ./convert-yolov3-tiny.py yolov3-tiny.weights
/ggml/examples/yolo/./convert-yolov3-tiny.py:21: RuntimeWarning: overflow encountered in cast
  l0_weights = l0_weights.astype(np.float16)
yolov3-tiny.weights converted to yolov3-tiny.gguf

Then running yolo detection is a failure:

(.venv) root@bb47dd5d8081:/ggml/examples/yolo# ../../build/bin/yolov3-tiny -m yolov3-tiny.gguf -i dog.jpg
ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
ggml_cuda_init: found 1 CUDA devices:
  Device 0: NVIDIA GeForce GTX 1080 Ti, compute capability 6.1, VMM: yes
create_backend: using CUDA0 backend
gguf_init_from_file_impl: tensor 'l2_biases' has offset 4928, expected 10848
gguf_init_from_file_impl: failed to read tensor data
load_model: gguf_init_from_file() failed
main: failed to load model from 'yolov3-tiny.gguf'

Nevertheless I found another yolov3-tiny.weights source that does the job (sha1sum is the same now: 40f3c11883bef62fd850213bc14266632ed4414f)
https://sourceforge.net/projects/yolov3.mirror/files/v8/yolov3-tiny.weights/download

Are the weights from https://pjreddie.com/media/files/yolov3-tiny.weights corrupted ? Or is there another underlying issue in convert-yolov3-tiny.py ?

Thank you very much :)