modelcloud/gptqmodel is a Python toolkit that compresses the weights of large language models (LLMs) to a low bit width so they can run with hardware-accelerated inference. The repository was created on June 17, 2024, and the latest version according to GitHub releases is v7.6.0. However, the given record does not include a list of changes for v7.6.0, so what differs cannot be confirmed. The source of the version information is the repository's GitHub releases page.
The README describes this repository as an extensible platform for LLM quantization, validation, and deployment. It bundles calibration, quantization, quality evaluation, model conversion, and accelerated inference into a single API, and covers formats such as GPTQ, AWQ, ParoQuant, GGUF, FP8, EXL3, and QQQ. Hardware acceleration targets NVIDIA CUDA, AMD ROCm, Huawei Ascend, Intel XPU, and Intel/AMD/Apple CPUs, and it is used together with Transformers, vLLM, and SGLang. The GitHub star count is 1274.
However, the license identifier remains NOASSERTION, so the reuse and distribution terms cannot be confirmed from the given record alone. The freshness, technology, and quality scores are not calculated and only the momentum score is 0.0, so the basis for judgment stays limited to the README description and the star count. There are also no figures on quantization loss, so the actual level of loss can only be guessed by comparing the original model and its output.