Modelpublic

liodon-ai/bloomz-560m-imatrix-GGUF

sourceHugging Facebigscience-bloom-rail-1.0updated 3mo agoView on Hugging Face

0likes227downloads

Model Card

bloomz-560m — iMatrix GGUF

GGUF quantizations of bigscience/bloomz-560m, published by Liodon AI.

Quick Start

llama.cpp

bash

llama-cli -hf liodon-ai/bloomz-560m-imatrix-GGUF:Q4_K_M

Ollama

bash

ollama run hf.co/liodon-ai/bloomz-560m-imatrix-GGUF:Q4_K_M

LM Studio / Jan — search liodon-ai/bloomz-560m-imatrix-GGUF and pick your quant.

Quants

Quant	Size	VRAM est.	Notes
`IQ2_M`	0.30 GB	~0 GB	2-bit, iMatrix — smallest usable
`IQ3_M`	0.37 GB	~0 GB	3-bit, iMatrix — great quality/size tradeoff
`IQ4_XS`	0.39 GB	~0 GB	4-bit extra-small, iMatrix
`Q4_K_M`	0.42 GB	~0 GB	4-bit, iMatrix-calibrated (recommended)
`Q5_K_M`	0.44 GB	~1 GB	5-bit, iMatrix-calibrated
`Q6_K`	0.47 GB	~1 GB	6-bit, iMatrix-calibrated, near-lossless
`Q8_0`	0.61 GB	~1 GB	8-bit, essentially lossless

What is iMatrix?

Standard quantization treats all weights equally. iMatrix runs 128 calibration chunks through the full-precision model to find which weights matter most, then allocates more precision where it counts. At Q2/Q3/Q4 this means noticeably better coherence and instruction-following — same file size, better output.

Calibration: 2M tokens of WikiText-103.

Also see plain (non-iMatrix) quants: liodon-ai/bloomz-560m-GGUF

Source

—Model: bigscience/bloomz-560m
—License: other

Quantized by [Liodon AI](https://huggingface.co/liodon-ai)