Standard
GGUF is a binary file format for storing quantized large language model weights, used by llama.cpp and compatible tools to run models on consumer hardware.
TornadoVM's University of Manchester team says its jitLLM engine runs LLM inference on NVIDIA GPUs from Java at about 90% of llama.cpp's speed. The figure comes from the project's own RTX 5090 recordings, so a JVM team still needs a run on its own GPUs before it retires a Python sidecar.
Reality
VIDRAFT released a 111GB, 4-bit build of its 180B-parameter Darwin model that it says runs on a laptop with 8GB of VRAM and 32GB of RAM. Air-gapped teams still need a measured token rate on that laptop before they can price it against a server.
Reality
Ollama and llama.cpp added local endpoints within a week for Jev decision models, 144M-to-27B checkpoints that return a label and a probability. Routing calls can move off hosted models once teams check each checkpoint's licence and test it on their own labels.
Reality
Salvatore Sanfilippo's open-source ds4 engine runs a short list of large open-weight models locally by compressing their routed experts to about two bits. Even compressed, the supported builds need high-memory Macs or GPU systems that most people do not already own.
Reality
VIDRAFT says POCKET-Darwin-180B, a 111 GB 4-bit GGUF build of its 180B mixture-of-experts model, runs in llama.cpp on about $1,400 of consumer hardware. The accuracy evidence so far is one MMLU-Pro comparison that VIDRAFT reports itself.
Reality
Prism's ternary Bonsai 2 27B ran at two to three tokens a second on a CPU-only Hetzner VPS in a dev.to test, against Simon Willison's 20 to 44 on a Mac. The sub-6GB file fits a 16GB box easily, but at that speed it only suits batch jobs.
Reality
Llama 8B on an RTX 4060 Ti 16GB fell from 42.5 to 3.8 tokens a second with a fifth of the model in system RAM, according to a dev.to benchmark. For a local coding agent, that makes VRAM for weights plus context the first spec to check, ahead of bandwidth.
Reality
Developer xbill9 rebuilt Google's QAT Gemma 4 26B as int4 and fit it on one TPU v6e with 53,888 tokens of KV cache, against 3,456 for RedHat's FP8 build. Throughput nearly doubles, with accuracy checked on one classification suite.
Reality
LM Studio 0.4.0 shipped llmster, a headless daemon that runs it on the Linux GPU servers where MIT-licensed Ollama already worked. Teams whose policy demands auditable source now decide on licence, since only LM Studio's lms CLI carries an MIT grant.
Reality
Ternary weights at 1.76 bits put a 27B model into a 5.9GB file and let a laptop decode it at 28.1 tokens a second. The retention figure comes from Prism ML's own benchmark suite, not the table on the model card.
Reality
The demo uploads a model file as base64 chunks over GET, then starts an inference server on it. It shows why egress and WAF rules keyed on the HTTP verb miss what the URL is doing.
Reality
PrismML's ternary build of Qwen3.8 27B keeps 98.2 percent of the full-precision benchmark average on both of the company's inconsistent scorecards, and the loss it does take is concentrated in knowledge and reasoning.
Reality
Every local runner now reads the same GGUF file, so the binding decision is the quant tag and the gigabyte or two of context that has to fit beside it. Ollama sets the GPU offload itself; llama.cpp lets you set it.
Reality
A dev.to walkthrough argues that the PROCESSOR field and the startup inference compute log line narrow a slow Ollama to about five named causes, and each of the three states PROCESSOR can print points at a different fix.
Reality
Two arms on the same laptop differ by one flag. The small card wins because llama.cpp leaves Gemma 4's 1.93 GB per-layer embedding table in mmap and pulls a few rows per token, so only about 1.08 GB of body is resident.
Reality
Alibaba's 27B scores 52 on the Artificial Analysis index from a 17GB quantized file. Filling its 262,144-token window needs roughly 16 GiB of KV cache on top of that, so the file size is the smaller half of the sizing question.
Reality
A dev.to guide puts the llama.cpp and Ollama decision on ownership. Under Ollama the named model is the unit you operate; under llama.cpp it is the llama-server process and the seven flags in its command line.
Reality
ROCm sits under PyTorch, vLLM and SGLang; Vulkan is what llama.cpp-class engines compile shaders against. The engine you already run narrows the choice to one, and the rest of the work is reading logs.
Reality
Per-tensor layout maps now drive his GGUF releases. The sensitivity data behind them came out of more than 1,000 quantizations of two Qwen3.5 models, scored by KL divergence against bf16 on wikitext-2-raw.
Reality
A 15M-parameter model streams English text on a 2007 PSP at about one token per second. That is the extreme end of a sizing rule. The harder half of that rule is checking whether the file that fits is a format its own maintainer recommends.
Reality
Build · August 20, 2026 · One report1 publisher