Enter the URI or path to the model file you want to import
huggingface://TheBloke/Llama-2-7B-Chat-GGUF
Standard HuggingFace format
hf://TheBloke/Llama-2-7B-Chat-GGUF
Short HuggingFace format
https://huggingface.co/TheBloke/Llama-2-7B-Chat-GGUF
Full HuggingFace URL
https://example.com/model.gguf
Direct download from any HTTPS URL
file:///path/to/model.gguf
Local file path (absolute)
/path/to/model.yaml
Direct local YAML config file
oci://registry.example.com/model:tag
OCI container registry
ocifile:///path/to/image.tar
Local OCI tarball file
ollama://llama2:7b
Ollama model format
https://example.com/model.yaml
Remote YAML config file
file:///path/to/config.yaml
Local YAML config file
Tip: For HuggingFace models, you can use any of the three formats. The system will automatically detect and download the appropriate model files.
Force a specific backend. Leave empty to auto-detect from URI.
Custom name for the model. If empty, the filename will be used.
Custom description for the model. If empty, a default description will be generated.
Preferred quantizations (comma-separated). Examples: q4_k_m, q4_k_s, q3_k_m, q2_k. Leave empty to use default (q4_k_m).
Preferred MMProj quantizations (comma-separated). Examples: fp16, fp32. Leave empty to use default (fp16).
Enable embeddings support for this model.
Model type for transformers backend. Examples: AutoModelForCausalLM, SentenceTransformer, Mamba, MusicgenForConditionalGeneration. Leave empty to use default (AutoModelForCausalLM).
Pipeline type for diffusers backend. Examples: StableDiffusionPipeline, StableDiffusion3Pipeline, FluxPipeline. Leave empty to use default (StableDiffusionPipeline).
Scheduler type for diffusers backend. Examples: k_dpmpp_2m, euler_a, ddim. Leave empty to use model default.
Enabled parameters for diffusers backend (comma-separated). Leave empty to use default (negative_prompt,num_inference_steps).
Enable CUDA support for GPU acceleration with diffusers backend.
Add custom key-value pairs for advanced configuration