List your company

Language Generation on C4

5.52PerplexityBaseline
5.445.986.527.06Oct 16, 2024Jan 19, 2025Apr 25, 2025Jul 30, 2025Nov 3, 2025Feb 7, 2026May 14, 2026

Evaluation results

MethodLinks
BaselinePrec.=FP16, Model=2-70B2026.055.52
OSAQ+GPTQPrec.=W4A16, Model=2-70B2026.055.61
-#Bits=W16A16, Model=Llama-65B2026.035.62
OmniQuantPrec.=W4A16, Model=2-70B2026.055.65
OSAQ+AWQPrec.=W4A16, Model=2-70B2026.055.66
GPTQPrec.=W4A16, Model=2-70B2026.055.67
SliderQuant#Bits=W4A16, Model=Llama-65B2026.035.7
AWQPrec.=W4A16, Model=2-70B2026.055.71
CBQ#Bits=W4A16, Model=Llama-65B2026.035.72
MagRPrec.=W4A16, Model=2-70B2026.055.72
OSAQ+AWQPrec.=W3A16 g128, Model=2-70B2026.055.72
OmniQuant#Bits=W4A16, Model=Llama-65B2026.035.73
AWQ#Bits=W4A16, Model=Llama-65B2026.035.77
OSAQ+GPTQPrec.=W3A16 g128, Model=2-70B2026.055.78
RTNPrec.=W4A16, Model=2-70B2026.055.79
GPTQ#Bits=W4A16, Model=Llama-65B2026.035.8
AWQPrec.=W3A16 g128, Model=2-70B2026.055.82
RTN#Bits=W4A16, Model=Llama-65B2026.035.85
OmniQuantPrec.=W3A16 g128, Model=2-70B2026.055.85
GPTQPrec.=W3A16 g128, Model=2-70B2026.055.85
QuIPPrec.=W4A16, Model=2-70B2026.055.87
MagRPrec.=W3A16 g128, Model=2-70B2026.056
RTNPrec.=W3A16 g128, Model=2-70B2026.056.02
OmniQuantPrec.=W3A16, Model=2-70B2026.056.06
OSAQ+QuIPPrec.=W3A16, Model=2-70B2026.056.06
OSAQ+GPTQPrec.=W3A16, Model=2-70B2026.056.09
QuIPPrec.=W3A16, Model=2-70B2026.056.14
BaselinePrec.=FP16, Model=2-13B2026.056.46
FP16Avg. Bit=16, Backbone=LLaMA2-13B2024.106.47
CMPQAvg. Bit=4, Backbone=LLaMA2-13B2024.106.55
AWQAvg. Bit=4, Backbone=LLaMA2-13B2024.106.56
GPTQPrec.=W3A16, Model=2-70B2026.056.57
LLM-MQAvg. Bit=4, Backbone=LLaMA2-13B2024.106.58
QuIP#Avg. Bit=4, Backbone=LLaMA2-13B2024.106.59
VPTQAvg. Bit=4, Backbone=LLaMA2-13B2024.106.59
-#Bits=W16A16, Model=Llama-13B2026.036.61
OSAQ+GPTQPrec.=W4A16, Model=2-13B2026.056.64
OmniQuantPrec.=W4A16, Model=2-13B2026.056.65
OSAQ+AWQPrec.=W4A16, Model=2-13B2026.056.67
GPTQPrec.=W4A16, Model=2-13B2026.056.7
SliderQuant#Bits=W4A16, Model=Llama-13B2026.036.73
AWQPrec.=W4A16, Model=2-13B2026.056.74
GPTQAvg. Bit=4, Backbone=LLaMA2-13B2024.106.74
OmniQuant#Bits=W4A16, Model=Llama-13B2026.036.76
CBQ#Bits=W4A16, Model=Llama-13B2026.036.77
MagRPrec.=W3A16, Model=2-70B2026.056.77
MagRPrec.=W4A16, Model=2-13B2026.056.81
RTNPrec.=W4A16, Model=2-13B2026.056.83
RTNAvg. Bit=4, Backbone=LLaMA2-13B2024.106.83
QuIPAvg. Bit=4, Backbone=LLaMA2-13B2024.106.83
GPTQ#Bits=W4A16, Model=Llama-13B2026.036.84
QuIPPrec.=W4A16, Model=2-13B2026.056.88
OSAQ+AWQPrec.=W3A16 g128, Model=2-13B2026.056.89
BaselinePrec.=FP16, Model=3-70B2026.056.9
CMPQAvg. Bit=3, Backbone=LLaMA2-13B2024.106.93
AWQPrec.=W3A16 g128, Model=2-13B2026.056.94
AWQAvg. Bit=3, Backbone=LLaMA2-13B2024.106.94
OSAQ+GPTQPrec.=W3A16 g128, Model=2-13B2026.056.95
BaselinePrec.=FP16, Model=2-7B2026.056.97
Full-precision#Bits=16/16, Size (GB)=13.542026.056.97
FP16Avg. Bit=16, Backbone=LLaMA2-7B2024.106.97
RTN#Bits=W4A16, Model=Llama-13B2026.036.98
AWQ#Bits=W4A16, Model=Llama-13B2026.036.98
OmniQuantPrec.=W3A16 g128, Model=2-13B2026.056.98
QuIP#Avg. Bit=3, Backbone=LLaMA2-13B2024.106.98
GPTQPrec.=W3A16 g128, Model=2-13B2026.057
VPTQAvg. Bit=3, Backbone=LLaMA2-13B2024.107.03
GPTVQAvg. Bit=3, Backbone=LLaMA2-13B2024.107.05
GPTQAvg. Bit=3, Backbone=LLaMA2-13B2024.107.06
-#Bits=W16A16, Model=Llama-7B2026.037.08
QuIPPrec.=W4A16, Model=3-70B2026.057.1
CMPQAvg. Bit=4, Backbone=LLaMA2-7B2024.107.1
AWQPrec.=W3A16, Model=2-70B2026.057.11
MagRPrec.=W3A16 g128, Model=2-13B2026.057.12
QuIPAvg. Bit=4, Backbone=LLaMA2-7B2024.107.12
AWQAvg. Bit=4, Backbone=LLaMA2-7B2024.107.13
VPTQAvg. Bit=4, Backbone=LLaMA2-7B2024.107.13
QuIP#Avg. Bit=4, Backbone=LLaMA2-7B2024.107.17
RTNPrec.=W3A16 g128, Model=2-13B2026.057.18
OSAQ+AWQPrec.=W4A16, Model=3-70B2026.057.22
LLM-MQAvg. Bit=4, Backbone=LLaMA2-7B2024.107.22
GPTQAvg. Bit=4, Backbone=LLaMA2-7B2024.107.23
OSAQ+GPTQPrec.=W4A16, Model=3-70B2026.057.24
QuIPAvg. Bit=3, Backbone=LLaMA2-13B2024.107.25
GPTVQAvg. Bit=4, Backbone=LLaMA2-7B2024.107.25
SliderQuant#Bits=W4A16, Model=Llama-7B2026.037.26
GPTVQAvg. Bit=4, Backbone=LLaMA2-13B2024.107.26
OSAQ+QuIPPrec.=W3A16, Model=2-13B2026.057.27
CBQ#Bits=W4A16, Model=Llama-7B2026.037.33
OmniQuant#Bits=W4A16, Model=Llama-7B2026.037.34
OSAQ+GPTQPrec.=W4A16, Model=2-7B2026.057.34
QuIPPrec.=W3A16, Model=2-13B2026.057.34
OmniQuantPrec.=W4A16, Model=2-7B2026.057.35
GPTQPrec.=W4A16, Model=2-7B2026.057.37
AWQPrec.=W4A16, Model=3-70B2026.057.4
GPTQPrec.=W4A16, Model=3-70B2026.057.4
GPTQ#Bits=W4A16, Model=Llama-7B2026.037.43
OmniQuantPrec.=W3A16, Model=2-13B2026.057.44
OSAQ+AWQPrec.=W4A16, Model=2-7B2026.057.5
AWQ#Bits=W4A16, Model=Llama-7B2026.037.52