{"slug":"huggingface-model-to-gguf","question":"How do I turn a HuggingFace model into a GGUF file?","answer":"Pull the model from the Hub, then convert and quantize it to GGUF with llama.cpp. a2a.computer plans this as a two-hop governed path and can realize each hop with an evidence receipt.","source":"hub_repo","target":"gguf","family":"Model embodiment","methodology":"The path below is produced live by the transmutation-planner over the real capability graph — each hop is a governed CHP capability, not a suggestion. Loss and cost are the planner's declared edge estimates; realizing a hop emits a signed evidence record.","limitations":["Quantization is lossy — the planner reports the expected quality loss; pick the quant (Q4_K_M, Q8_0, …) to trade size against fidelity.","Cost and latency are static estimates, not per-host measurements (per-host figures are a planned enrichment)."],"related":["fine-tuned-model-to-gguf","audio-to-image"],"updated":"2026-08-12","paths":{"source":"hub_repo","target":"gguf","paths":[{"rank":1,"score":0.5,"total_loss":0.15,"total_cost_s":300,"hops":2,"max_risk":"high","steps":[{"capability_id":"chp.adapters.huggingface.pull","consumes":"hub_repo","produces":"hf_dir","loss":0,"cost_s":120,"risk":"medium","note":"Download a Hub model repo to a local HF directory."},{"capability_id":"chp.adapters.huggingface.quantize_to_gguf","consumes":"hf_dir","produces":"gguf","loss":0.15,"cost_s":180,"risk":"high","note":"Convert + quantize an HF model to GGUF (llama.cpp; Q4_K_M/Q8_0/...)."}]}],"eliminated":[],"diagnostic":null},"path_error":null,"capability_ids":["chp.adapters.huggingface.pull","chp.adapters.huggingface.quantize_to_gguf"],"mcp_tools":["find_transmutation_paths","estimate_transmutation","realize_state","retrieve_evidence"]}