← HomeGUIDE 03 / COMPATIBILITY

Supported GGUF models and compatibility

Understand which llama.cpp-compatible GGUF models Local Idea Studio can load, how to check model compatibility, and why RAM, VRAM, and architecture matter.

What Local Idea Studio supports

Local Idea Studio is designed for single-file, llama.cpp-compatible GGUF text-generation models. Common families such as Llama, Qwen, Mistral, Gemma, Phi, DeepSeek, Code Llama, and TinyLlama commonly have compatible GGUF releases, but compatibility is determined by the exact architecture and runtime version—not only the repository name.

Compatibility checklist

  • The downloaded file ends in .gguf.
  • The repository describes the model as compatible with llama.cpp.
  • The file is a complete model, not one shard from a split set.
  • The architecture is supported by the bundled runtime.
  • The model plus context fits available RAM or VRAM.
  • The model’s license allows your intended use.

Backends

BackendHardwareUse it when
CUDASupported NVIDIA RTX/GTXAuto detects NVIDIA CUDA and normally prefers it for performance.
VulkanSupported AMD, NVIDIA, and Intel GPUsUse the cross-vendor GPU path when CUDA is unavailable or for non-NVIDIA hardware.
CPUModern x64 processorUse for compatibility, limited VRAM, or comparison testing.

What is not promised

Local Idea Studio does not guarantee that every file labelled GGUF will load. Very new architectures can require a newer llama.cpp build, multimodal models may need additional projector files or runtime support, and unusually large contexts can exceed memory even when model weights fit.

If a model is rejected

Record the exact Hugging Face repository, filename, quantization, file size, error text, backend, context size, and hardware. Search existing GitHub issues before opening a report. Do not include private tokens, chats, or personal paths.