GGUF vs GPTQ vs AWQ vs EXL2: LLM Model Formats Explained (2026)

來源原標題

GGUF vs GPTQ vs AWQ vs EXL2: LLM Model Formats Explained (2026)

作者

Asif Razzaq

原始標籤

Agentic AI · AI Governance · AI Infrastructure · AI Shorts · Artificial Intelligence · Comparison · Deep Learning · Editors Pick · For Devs · Language Model · Large Language Model · Machine Learning · Staff · Tech News · Technology

原始連結

https://www.marktechpost.com/2026/09/18/gguf-vs-gptq-vs-awq-vs-exl2-llm-model-formats-explained-2026/

摘要原文

Most confusion comes from mixing 2 layers. A container defines how tensors are stored on disk. A quantization method defines how weights are squeezed into fewer bits. Weight memory ≈ parameters × bits-per-weight ÷ 8. This is arithmetic, not a vendor benchmark. It covers weights only. The KV cache and runtime overhead add more on top. Unquantized models usually ship as 16-bit weights, in either pytorch_model.bin or model.safetensors . The older .bin / .pt files use Python pickle. Loading a pickle file can execute arbitrary code, which makes untrusted checkpoints a security risk. Safetensors , created at Hugging Face, removes that risk. A file is a small JSON header plus raw tensor buffers, with nothing executable inside. Tensors can be memory-mapped and loaded one at a time without reading the whole file. Safetensors is now listed as a PyTorch Foundation project . Important nuance: most GPTQ, AWQ, EXL2, EXL3, and MLX models are also stored in .safetensors files. The quantization lives in the tensor contents and a config file, not in a new container. GGUF is a binary format for running models with GGML and GGML-based executors such as llama.cpp. It was created by Georgi Gerganov, who also leads llama.cpp ( Hugging Face docs ). It was introduced on August 21, 2023 as the replacement for the older GGML format. The older GGML, GGMF, and GGJT files could not say which architecture a model belonged to. Adding a new hyperparameter broke every existing file. GGUF switched to typed key-value metadata, so new fields can be added without breaking old files. The spec lists 5 goals: single-file deployment, extensibility, mmap compatibility, easy loading, and complete information inside the file. Unlike tensor-only formats, GGUF can carry the tokenizer, special tokens, and a Jinja chat template alongside the weights. The suffix in a name like Q4_K_M.gguf tells you the scheme. Figures below come from the Hugging Face GGUF docs . *Derived by hand, not listed in the HF table: 32 weights plus a 16-bit scale (and a 16-bit minimum for Q4_1). Checking the Q4_K math : A super-block holds 256 weights. 256 × 4 bits = 1,024 bits. Add 8 blocks × 12 bits of scales and minimums (96 bits). Add a 16-bit super-scale and 16-bit super-minimum (32 bits). Total: 1,152 ÷ 256 = 4.5 bits per weight. What _S , _M , _L mean : These are mixes, not new types. For example, llama.cpp describes Q4_K_M as using Q6_K for half of the attention.wv and feed_forward.w2 tensors and Q4_K elsewhere ( Unsloth docs ). That is why a Q4_K_M file averages above 4.5 bits per weight. Newer types : The HF table also lists TQ1_0 and TQ2_0 for ternary weights, plus MXFP4, a 4-bit microscaling floating-point type. A labeling quirk : Hugging Face files Q8_0 under “legacy” types. In practice, Q8_0 remains the standard near-lossless GGUF choice. Hugging Face’s reference table for a Llama-2-7B-class model shows the trade-off : Illustrative only. These numbers come from a 2023-era 7B model; newer models can react differently. GGUF quantization can use calibration data. llama.cpp’s llama-imatrix computes an importance matrix from a text file. llama-quantize --imatrix then uses it to improve quality. For 1-bit and 2-bit mixes, llama-quantize warns if no imatrix is supplied . The spec defines filenames as base name, size label, fine-tune, version, encoding, type, and shard. Shards use a 5-digit counter such as 00003-of-00009 . Optional mmproj- and mtp- prefixes mark vision projectors and multi-token-prediction draft modules. GGUF is native to llama.cpp and its ecosystem. Hugging Face documents use with llama.cpp, LM Studio, GPT4All, and Ollama . vLLM support exists but is limited. vLLM calls it highly experimental and under-optimized, and GGUF now needs the out-of-tree vllm-gguf-plugin . GPTQ was written by Elias Frantar (IST Austria), Saleh Ashkboos and Torsten Hoefler (ETH Zurich), and Dan Alistarh (IST Austria & Neural Magic). It first appeared on arXiv on October 31, 2022. It was published at ICLR 2023 . GPTQ is a one-shot, post-training weight quantization method. It uses approximate second-order (Hessian) information to decide how to round weights. Rounding error in one column is compensated by adjusting weights not yet quantized. It needs a small calibration dataset but no retraining. GPTQ repos often include GPTQ or tags like 4bit-128g in the name. The original AutoGPTQ library is no longer maintained . AWQ (Activation-aware Weight Quantization) comes from Song Han’s group at MIT. It first appeared on arXiv on June 1, 2023. It won the MLSys 2024 Best Paper Award . Not all weights matter equally. Protecting roughly 1% of ‘salient’ weights sharply reduces quantization error . The twist: AWQ finds those salient channels by looking at activation magnitudes, not the weights themselves. It does not store those channels at higher precision. Instead, it scales them up through a mathematically equivalent transformation, keeping a uniform, hardware-friendly format. AWQ uses no backpropagation or reconstruction, so it is less likely to overfit its calibration set. EXL2 is the native format of ExLlamaV2 , an inference library by turboderp for consumer GPUs. It uses the same optimization method as GPTQ and supports 2, 3, 4, 5, 6, and 8-bit quantization. That is why EXL2 files carry names like 4.65bpw instead of 4-bit . TabbyAPI is the official recommended server, providing an OpenAI-compatible API. EXL2 renames some tensors so every model looks like a Llama variant internally . That makes EXL2 hard to reuse in other frameworks. EXL3 is the successor format, built on QTIP from Cornell RelaxML. QTIP uses trellis-coded quantization with incoherence processing and was published at NeurIPS 2024. EXL3 keeps QTIP’s procedural codebook and trellis encoding. It changes how tensors are regularized and packed. ExLlamaV3 adds 2–8 bit KV-cache quantization, tensor- and expert-parallel inference, speculative decoding, multimodal support, and a Transformers plugin. Recent releases add CPU offloading for large MoE models . Hardware note: ExLlamaV3 requires CUDA 12.4 or later. Its README has listed ROCm support as a to-do item. bitsandbytes is usually not something you download pre-quantized. You load a 16-bit model and quantize it on the fly. Its 4-bit mode comes from QLoRA : MLX-LM is a Python package for running and fine-tuning LLMs on Apple Silicon with MLX. MLX comes from Apple Machine Learning Research . MLX models are safetensors with MLX-specific quantized weights. mlx_lm.convert with -q quantizes a Hugging Face model and can upload it to the mlx-community organization. On a Mac, both GGUF (via llama.cpp) and MLX are strong options. The post GGUF vs GPTQ vs AWQ vs EXL2: LLM Model Formats Explained (2026) appeared first on MarkTechPost .