ResearchBenchmarksModel Compression on Llama 8B 3.1Model Compression on Llama 8B 3.1Follow2.8Model Size (GB)UniQL2.2725.8369.412.964Dec 3, 2025Evaluation resultsMethodMethodLinksModel Size (GB)UniQLW-bit=4, Prun. p%=35%W-bit=4, Prun. p%=35%2025.122.8UniQLW-bit=4, Prun. p%=0%W-bit=4, Prun. p%=0%2025.124.1TAO-HQQW-bit=4*, Prun. p%=0%W-bit=4*, Prun. p%=0%2025.125.7TRT-AWQW-bit=4*, Prun. p%=0%W-bit=4*, Prun. p%=0%2025.125.8MoDeGPTW-bit=16, Prun. p%=15%W-bit=16, Prun. p%=15%2025.1213.9SVD-LLMW-bit=16, Prun. p%=15%W-bit=16, Prun. p%=15%2025.1214.1FP16W-bit=16, Prun. p%=0%W-bit=16, Prun. p%=0%2025.1216