# Low-memory optimization

This is the Llamacpp imatrix quantized version of the Mistral-Small-3.2-24B-Instruct-2506 model, offering various quantization types to meet different hardware requirements.

This is an experimental small reasoning model designed to run on 8GiB consumer-grade GPUs with general inference capabilities. Through supervised fine-tuning (SFT) and high-quality reasoning trajectory training, the model can generalize its reasoning abilities to multiple tasks.

Qwen3-14B-abliterated is a quantized version based on the Qwen3-14B model, optimized using llama.cpp, offering multiple quantization options to meet different performance requirements.

Quantized version based on Qwen/Qwen3-32B, using llama.cpp for quantization, supporting multiple quantization types for different hardware requirements.

Llama 2 is a 7B-parameter large language model developed by Meta, offering multiple quantization versions to accommodate different hardware requirements.

phi-4 is an open-source language model developed by Microsoft Research, focusing on high-quality data and reasoning capabilities, suitable for memory/computation-constrained environments.

RWKV-7 model based on flash-linear attention format, supporting multilingual text generation tasks.

Cydonia-24B-v2.1 is a 24B parameter large language model, processed with llama.cpp's imatrix quantization, offering multiple quantized versions to suit different hardware requirements.

Rombo-LLM-V3.1-QWQ-32b is a 32B-parameter large language model, processed with llama.cpp's imatrix quantization, offering multiple quantization versions to accommodate different hardware requirements.

Llamacpp imatrix quantized version of Nera_Noctis-12B, based on Nitral-AI/Nera_Noctis-12B model, supporting English text generation tasks.

Aura-4B is a quantized version based on AuraIndustries/Aura-4B, using llama.cpp for imatrix quantization, supporting multiple quantization types, suitable for text generation tasks.