Cannot import some HF models

[dinccey]

LocalAI version:
Docker localai/localai:latest-nvidia-l4t-arm64-cuda-13 v4.7.1

Environment, CPU architecture, OS, and Version:
ARM64 nvidia ubuntu
Describe the bug
Trying to import the new deepseek model from unsloth doesn't work. Advanced import hf://unsloth/DeepSeek-V4-Flash-0731-GGUF and quantization UD-IQ3_XXS yields this error:

Failed to install model: DeepSeek-V4-Flash-0731-GGUF(could not fetch "/models/ds4flash.gguf": local file does not exist (stat /models/ds4flash.gguf: no such file or directory) and "hf://unsloth/DeepSeek-V4-Flash-0731-GGUF" is not a recognized downloadable URL (supported schemes: http://, https://, file://, huggingface://, hf://, ollama://, oci://, ocifile://, github://))
To Reproduce
try to import unsloth/DeepSeek-V4-Flash-0731-GGUF
Expected behavior
model is downloaded correctly and can be loaded with DS4 or other backends
Logs
Failed to install model: DeepSeek-V4-Flash-0731-GGUF(could not fetch "/models/ds4flash.gguf": local file does not exist (stat /models/ds4flash.gguf: no such file or directory) and "hf://unsloth/DeepSeek-V4-Flash-0731-GGUF" is not a recognized downloadable URL (supported schemes: http://, https://, file://, huggingface://, hf://, ollama://, oci://, ocifile://, github://))

[localai-org-maint-bot]

I’m picking this up. I’ll reproduce the DS4 Hugging Face repository import selecting the repo URI instead of the requested quantization, add focused regression coverage, and make the smallest importer fix on current master.