List your company

Language Modeling on Perplexity

3.43Perplexity (PPL)SERQ (GPTQ)
3.06645.52077.97510.4293Nov 11, 2025Dec 2, 2025Dec 23, 2025Jan 14, 2026Feb 4, 2026Feb 25, 2026Mar 19, 2026

Evaluation results

MethodLinks
SERQ (GPTQ)Backbone=LLaMA-2 70B, #Bits=W4A8, #Eff. (w)=3.432026.033.43
SERQ (RTN)Backbone=LLaMA-2 70B, #Bits=W4A8, #Eff. (w)=3.442026.033.44
baselineBackbone=LLaMA-2 70B, #Bits=FP16, #Eff. (w)=162026.033.53
L2QERBackbone=LLaMA-2 70B, #Bits=W4A8, #Eff. (w)=3.552026.033.55
SERQ (GPTQ)Backbone=LLaMA-2 70B, #Bits=W4A4, #Eff. (w)=3.642026.033.64
SERQ (RTN)Backbone=LLaMA-2 70B, #Bits=W4A4, #Eff. (w)=3.652026.033.65
SERQ-MXFP4Backbone=LLaMA-2 70B, #Bits=W4A4, #Eff. (w)=3.792026.033.79
L2QER-MXFP4Backbone=LLaMA-2 70B, #Bits=W4A4, #Eff. (w)=3.812026.033.81
L2QERBackbone=LLaMA-2 70B, #Bits=W4A4, #Eff. (w)=4.552026.034.55
Clean FT (No WM)Model=Llama-2-7B, Condition=Original (Pre-Attack)2026.035.3
baselineBackbone=LLaMA-2 7B, #Bits=FP16, #Eff. (w)=162026.035.47
Pre-trainedModel Backbone=LLaMA2-7B, Precision=FP162025.115.47
SERQ (GPTQ)Backbone=LLaMA-2 7B, #Bits=W4A8, #Eff. (w)=4.242026.035.59
SERQ (RTN)Backbone=LLaMA-2 7B, #Bits=W4A8, #Eff. (w)=4.242026.035.64
EmMarkModel=Llama-2-7B, Condition=Attacked (Post-Attack)2026.035.73
Weight Quant.Model=Llama-2-7B, Condition=Attacked (Post-Attack)2026.035.78
L2QERBackbone=LLaMA-2 7B, #Bits=W4A8, #Eff. (w)=4.352026.035.83
Clean FT (No WM)Model=Llama-3-8B, Condition=Original (Pre-Attack)2026.035.83
Full DIFSWModel=Llama-2-7B, Condition=Original (Pre-Attack)2026.035.91
Full DIFSWModel=Llama-2-7B, Condition=Attacked (Post-Attack)2026.035.91
SERQ (GPTQ)Backbone=LLaMA-2 7B, #Bits=W4A4, #Eff. (w)=4.242026.035.97
SERQ (RTN)Backbone=LLaMA-2 7B, #Bits=W4A4, #Eff. (w)=4.242026.036.03
Weight Quant.Model=Llama-3-8B, Condition=Attacked (Post-Attack)2026.036.03
EmMarkModel=Llama-3-8B, Condition=Attacked (Post-Attack)2026.036.09
Full DIFSWModel=Llama-3-8B, Condition=Attacked (Post-Attack)2026.036.12
baselineBackbone=LLaMA-3 8B, #Bits=FP16, #Eff. (w)=162026.036.13
Pre-trainedModel Backbone=LLaMA3.1-8B, Precision=FP162025.116.14
Full DIFSWModel=Llama-3-8B, Condition=Original (Pre-Attack)2026.036.2
SERQ-MXFP4Backbone=LLaMA-2 7B, #Bits=W4A4, #Eff. (w)=4.372026.036.22
Weight Quant.Model=Mistral-7B, Condition=Attacked (Post-Attack)2026.036.23
EmMarkModel=Mistral-7B, Condition=Attacked (Post-Attack)2026.036.24
L2QER-MXFP4Backbone=LLaMA-2 7B, #Bits=W4A4, #Eff. (w)=4.372026.036.3
Clean FT (No WM)Model=Qwen2.5-7B, Condition=Original (Pre-Attack)2026.036.35
SERQ (GPTQ)Backbone=LLaMA-3 8B, #Bits=W4A8, #Eff. (w)=6.522026.036.52
Full DIFSWModel=Mistral-7B, Condition=Attacked (Post-Attack)2026.036.59
Naive Top-kModel=Llama-3-8B, Condition=Attacked (Post-Attack)2026.036.64
Weight Quant.Model=Qwen2.5-7B, Condition=Attacked (Post-Attack)2026.036.66
EmMarkModel=Qwen2.5-7B, Condition=Attacked (Post-Attack)2026.036.68
SERQ (RTN)Backbone=LLaMA-3 8B, #Bits=W4A8, #Eff. (w)=6.712026.036.71
Naive Top-kModel=Llama-2-7B, Condition=Attacked (Post-Attack)2026.036.75
Naive Top-kModel=Mistral-7B, Condition=Attacked (Post-Attack)2026.036.93
Fine-tunedModel Backbone=LLaMA2-7B, Precision=FP162025.116.94
Pre-trainedModel Backbone=Qwen2-7B, Precision=FP162025.117.13
L2QERBackbone=LLaMA-3 8B, #Bits=W4A8, #Eff. (w)=7.162026.037.16
Fine-tunedModel Backbone=LLaMA3.1-8B, Precision=FP162025.117.23
Full DIFSWModel=Qwen2.5-7B, Condition=Original (Pre-Attack)2026.037.24
LLM.int4()Backbone=LLaMA-2 7B, #Bits=W4A8, #Eff. (w)=-†2026.037.28
OSTQuantModel Backbone=LLaMA2-7B, Precision=W4A42025.117.28
Naive Top-kModel=Llama-2-7B, Condition=Original (Pre-Attack)2026.037.3
Naive Top-kModel=Llama-3-8B, Condition=Original (Pre-Attack)2026.037.35
L2QERBackbone=LLaMA-2 7B, #Bits=W4A4, #Eff. (w)=4.242026.037.37
CAQModel Backbone=LLaMA2-7B, Precision=W4A42025.117.56
Fine-tunedModel Backbone=Qwen2-7B, Precision=FP162025.117.6
SERQ-MXFP4Backbone=LLaMA-3 8B, #Bits=W4A4, #Eff. (w)=7.632026.037.63
Full DIFSWModel=Qwen2.5-7B, Condition=Attacked (Post-Attack)2026.037.66
QuaRotModel Backbone=LLaMA2-7B, Precision=W4A42025.117.67
SpinQuantModel Backbone=LLaMA2-7B, Precision=W4A42025.117.68
SERQ (GPTQ)Backbone=LLaMA-3 8B, #Bits=W4A4, #Eff. (w)=7.752026.037.75
Naive Top-kModel=Qwen2.5-7B, Condition=Attacked (Post-Attack)2026.037.79
baselineBackbone=LLaMA-3.2 3B, #Bits=FP16, #Eff. (w)=162026.037.81
L2QER-MXFP4Backbone=LLaMA-3 8B, #Bits=W4A4, #Eff. (w)=7.832026.037.83
DuQuantModel Backbone=LLaMA2-7B, Precision=W4A42025.117.88
Naive Top-kModel=Qwen2.5-7B, Condition=Original (Pre-Attack)2026.037.94
EmMarkModel=DeepSeek-7B, Condition=Attacked (Post-Attack)2026.037.95
baselineBackbone=Qwen-2.5-3B, #Bits=FP16, #Eff. (w)=162026.038.03
SERQ (RTN)Backbone=LLaMA-3 8B, #Bits=W4A4, #Eff. (w)=8.072026.038.07
SERQ (GPTQ)Backbone=LLaMA-3.2 3B, #Bits=W4A8, #Eff. (w)=4.242026.038.18
OSTQuantModel Backbone=Qwen2-7B, Precision=W4A42025.118.18
CAQModel Backbone=Qwen2-7B, Precision=W4A42025.118.23
Weight Quant.Model=DeepSeek-7B, Condition=Attacked (Post-Attack)2026.038.23
OSTQuantModel Backbone=LLaMA3.1-8B, Precision=W4A42025.118.29
SERQ (GPTQ)Backbone=Qwen-2.5-3B, #Bits=W4A8, #Eff. (w)=4.242026.038.36
SERQ (RTN)Backbone=LLaMA-3.2 3B, #Bits=W4A8, #Eff. (w)=4.242026.038.38
CAQModel Backbone=LLaMA3.1-8B, Precision=W4A42025.118.41
SpinQuantModel Backbone=Qwen2-7B, Precision=W4A42025.118.41
QuaRotModel Backbone=Qwen2-7B, Precision=W4A42025.118.43
SpinQuantModel Backbone=LLaMA3.1-8B, Precision=W4A42025.118.57
SERQ (RTN)Backbone=Qwen-2.5-3B, #Bits=W4A8, #Eff. (w)=4.242026.038.67
SERQ-MXFP4Backbone=LLaMA-3.2 3B, #Bits=W4A4, #Eff. (w)=4.372026.038.74
L2QERBackbone=LLaMA-3.2 3B, #Bits=W4A8, #Eff. (w)=4.352026.038.78
QuaRotModel Backbone=LLaMA3.1-8B, Precision=W4A42025.118.81
SERQ (GPTQ)Backbone=LLaMA-3.2 3B, #Bits=W4A4, #Eff. (w)=4.242026.039.15
DuQuantModel Backbone=LLaMA3.1-8B, Precision=W4A42025.119.22
SERQ (GPTQ)Backbone=Qwen-2.5-3B, #Bits=W4A4, #Eff. (w)=4.242026.039.35
SERQ (RTN)Backbone=LLaMA-3.2 3B, #Bits=W4A4, #Eff. (w)=4.242026.039.43
L2QER-MXFP4Backbone=LLaMA-3.2 3B, #Bits=W4A4, #Eff. (w)=4.372026.039.59
SERQ (RTN)Backbone=Qwen-2.5-3B, #Bits=W4A4, #Eff. (w)=4.242026.039.66
baselineBackbone=LLaMA-3.2 1B, #Bits=FP16, #Eff. (w)=162026.039.75
Weight Quant.Model=Llama-3-8B, Condition=Original (Pre-Attack)2026.039.77
SERQ-MXFP4Backbone=Qwen-2.5-3B, #Bits=W4A4, #Eff. (w)=4.372026.039.79
SERQ (GPTQ)Backbone=LLaMA-3.2 1B, #Bits=W4A8, #Eff. (w)=4.242026.0310.45
L2QER-MXFP4Backbone=Qwen-2.5-3B, #Bits=W4A4, #Eff. (w)=4.372026.0310.46
LLM.int4()Backbone=LLaMA-3 8B, #Bits=W4A8, #Eff. (w)=-†2026.0310.82
SERQ (RTN)Backbone=LLaMA-3.2 1B, #Bits=W4A8, #Eff. (w)=4.242026.0311.14
L2QERBackbone=LLaMA-3 8B, #Bits=W4A4, #Eff. (w)=11.442026.0311.44
DensePrune%=0%, Evaluation=Zero-shot, Model=Mistral-7B2025.1211.9
L2QERBackbone=LLaMA-3.2 1B, #Bits=W4A8, #Eff. (w)=4.352026.0312.06
L2QERBackbone=Qwen-2.5-3B, #Bits=W4A8, #Eff. (w)=4.352026.0312.25
LLM.int4()Backbone=LLaMA-3.2 3B, #Bits=W4A8, #Eff. (w)=-†2026.0312.47
SERQ (GPTQ)Backbone=LLaMA-3.2 1B, #Bits=W4A4, #Eff. (w)=4.242026.0312.52