List your company
ResearchFormerly WizWand

Language Modeling on C4 (test)

4.97PerplexityFP16
4.893165.411835.93056.44917Feb 14, 2024Jun 23, 2024Oct 31, 2024Mar 10, 2025Jul 18, 2025Nov 25, 2025Apr 5, 2026
Updated 4mo ago

Evaluation results

MethodLinks
FP16Bits=16.0, Model=Llama-2 70B, Context Length=40962024.064.97
QTIPBits=4.0, Model=Llama-2 70B, Context Length=40962024.065
QuIP#Bits=4.0, Model=Llama-2 70B, Context Length=40962024.065.02
AQLMBits=4.0, Model=Llama-2 70B, Context Length=40962024.065.03
QTIPBits=3.0, Model=Llama-2 70B, Context Length=40962024.065.08
QuIP#Bits=3.0, Model=Llama-2 70B, Context Length=40962024.065.15
AQLMBits=3.0, Model=Llama-2 70B, Context Length=40962024.065.17
QTIPBits=2.0, Model=Llama-2 70B, Context Length=40962024.065.48
DuQuant#Bit=W6A6, Backbone=LLaMA1-30B2026.045.49
FP16Bits=16.0, Model=Llama-1 65B, Context Length=20482024.065.62
AQLMBits=2.0, Model=Llama-2 70B, Context Length=40962024.065.62
QTIPBits=4.0, Model=Llama-1 65B, Context Length=20482024.065.64
QuIP#Bits=4.0, Model=Llama-1 65B, Context Length=20482024.065.66
QTIPBits=3.0, Model=Llama-1 65B, Context Length=20482024.065.71
QuIP#Bits=2.0, Model=Llama-2 70B, Context Length=40962024.065.71
FP16Bits Per FPN=16, Backbone=Mistral-7b2024.055.71
KVQuant-4b-1%Bits Per FPN=4.32-4.35, Backbone=Mistral-7b2024.055.72
CQ-2c8bBits Per FPN=4.00, Backbone=Mistral-7b2024.055.74
KVQuant-4bBits Per FPN=4.00-4.02, Backbone=Mistral-7b2024.055.75
INT4-gs128Bits Per FPN=4.16, Backbone=Mistral-7b2024.055.76
QuIP#Bits=3.0, Model=Llama-1 65B, Context Length=20482024.065.77
NF4-gs128Bits Per FPN=4.16, Backbone=Mistral-7b2024.055.77
NF4Bits Per FPN=4.00-4.01, Backbone=Mistral-7b2024.055.85
INT4Bits Per FPN=4.00-4.01, Backbone=Mistral-7b2024.055.91
FP16#Bit=FP16, Backbone=LLaMA1-30B2026.045.92
RUQuant#Bit=W6A6, Backbone=LLaMA1-30B2026.045.94
FP16Bits=16.0, Model=Llama-1 30B, Context Length=20482024.065.98
QTIPBits=4.0, Model=Llama-1 30B, Context Length=20482024.066.01
QuIP#Bits=4.0, Model=Llama-1 30B, Context Length=20482024.066.03
FP16Bits=16.0, Model=Llama-2 13B, Context Length=40962024.066.05
FP16Bits Per FPN=16, Backbone=LLaMA-2-13b2024.056.05
KVQuant-4b-1%Bits Per FPN=4.32-4.35, Backbone=LLaMA-2-13b2024.056.06
QTIPBits=2.0, Model=Llama-1 65B, Context Length=20482024.066.08
KVQuant-2b-1%Bits Per FPN=2.32-2.35, Backbone=Mistral-7b2024.056.08
QTIPBits=4.0, Model=Llama-2 13B, Context Length=40962024.066.09
QTIPBits=3.0, Model=Llama-1 30B, Context Length=20482024.066.09
CQ-2c8bBits Per FPN=4.00, Backbone=LLaMA-2-13b2024.056.09
KVQuant-4bBits Per FPN=4.00-4.02, Backbone=LLaMA-2-13b2024.056.11
QuIP#Bits=4.0, Model=Llama-2 13B, Context Length=40962024.066.13
FP16Model Family=LLaMA, Model Scale=30B, Precision=FP162024.026.131
AQLMBits=4.0, Model=Llama-2 13B, Context Length=40962024.066.14
QuIP#Bits=3.0, Model=Llama-1 30B, Context Length=20482024.066.17
CQ-4c8bBits Per FPN=2.00, Backbone=Mistral-7b2024.056.17
QuIP#Bits=2.0, Model=Llama-1 65B, Context Length=20482024.066.19
INT4-gs128Bits Per FPN=4.16, Backbone=LLaMA-2-13b2024.056.2
NF4-gs128Bits Per FPN=4.16, Backbone=LLaMA-2-13b2024.056.2
QTIPBits=3.0, Model=Llama-2 13B, Context Length=40962024.066.22
RUQuant#Bit=W4A4, Backbone=LLaMA1-30B2026.046.25
aespaModel Family=LLaMA, Model Scale=30B, Precision=INT42024.026.256
DuQuant#Bit=W4A4, Backbone=LLaMA1-30B2026.046.26
OmniQuantModel Family=LLaMA, Model Scale=30B, Precision=INT42024.026.269
AffineQuantModel Family=LLaMA, Model Scale=30B, Precision=INT42024.026.27
AQLMBits=3.0, Model=Llama-2 13B, Context Length=40962024.066.33
QuIP#Bits=3.0, Model=Llama-2 13B, Context Length=40962024.066.35
KVQuant-2b-1%Bits Per FPN=2.32-2.35, Backbone=LLaMA-2-13b2024.056.38
FP16#Bit=FP16, Backbone=LLaMA2-13B2026.046.41
RUQuant#Bit=W6A6, Backbone=LLaMA2-13B2026.046.43
DuQuant#Bit=W6A6, Backbone=LLaMA2-13B2026.046.44
NF4Bits Per FPN=4.00-4.01, Backbone=LLaMA-2-13b2024.056.45
FP16#Bit=FP16, Backbone=LLaMA1-13B2026.046.49
RUQuant#Bit=W6A6, Backbone=LLaMA1-13B2026.046.51
CQ-4c8bBits Per FPN=2.00, Backbone=LLaMA-2-13b2024.056.52
DuQuant#Bit=W6A6, Backbone=LLaMA1-13B2026.046.52
QTIPBits=2.0, Model=Llama-1 30B, Context Length=20482024.066.56
INT4Bits Per FPN=4.00-4.01, Backbone=LLaMA-2-13b2024.056.59
FP16Bits=16.0, Model=Llama-1 13B, Context Length=20482024.066.61
FP16Bits Per FPN=16, Backbone=LLaMA-13b2024.056.61
KVQuant-4b-1%Bits Per FPN=4.32-4.35, Backbone=LLaMA-13b2024.056.62
FP16Bits=16.0, Model=Llama-2 7B, Context Length=40962024.066.63
FP16Bits Per FPN=16, Backbone=LLama-2-7b2024.056.63
CQ-2c8bBits Per FPN=4.00, Backbone=LLaMA-13b2024.056.64
QTIPBits=4.0, Model=Llama-1 13B, Context Length=20482024.066.65
KVQuant-4bBits Per FPN=4.00-4.02, Backbone=LLaMA-13b2024.056.65
KVQuant-4b-1%Bits Per FPN=4.32-4.35, Backbone=LLama-2-7b2024.056.65
NF4-gs128Bits Per FPN=4.16, Backbone=LLaMA-13b2024.056.66
QuIP#Bits=4.0, Model=Llama-1 13B, Context Length=20482024.066.67
INT4-gs128Bits Per FPN=4.16, Backbone=LLaMA-13b2024.056.67
CQ-2c8bBits Per FPN=4.00, Backbone=LLama-2-7b2024.056.67
QTIPBits=4.0, Model=Llama-2 7B, Context Length=40962024.066.69
KVQuant-4bBits Per FPN=4.00-4.02, Backbone=LLama-2-7b2024.056.7
QuIP#Bits=2.0, Model=Llama-1 30B, Context Length=20482024.066.71
FP16Model Family=LLaMA2, Model Scale=13B, Precision=FP162024.026.727
QTIPBits=3.0, Model=Llama-1 13B, Context Length=20482024.066.74
NF4Bits Per FPN=4.00-4.01, Backbone=LLaMA-13b2024.056.74
AQLMBits=4.0, Model=Llama-2 7B, Context Length=40962024.066.75
QuIP#Bits=4.0, Model=Llama-2 7B, Context Length=40962024.066.75
FP16Model Family=LLaMA, Model Scale=13B, Precision=FP162024.026.798
INT4Bits Per FPN=4.00-4.01, Backbone=LLaMA-13b2024.056.82
QuIP#Bits=3.0, Model=Llama-1 13B, Context Length=20482024.066.83
KVQuant-2b-1%Bits Per FPN=2.32-2.35, Backbone=LLaMA-13b2024.056.83
FP16#Bit=FP16, Backbone=LLaMA2-7B2026.046.84
QTIPBits=2.0, Model=Llama-2 13B, Context Length=40962024.066.85
NF4-gs128Bits Per FPN=4.16, Backbone=LLama-2-7b2024.056.86
QTIPBits=3.0, Model=Llama-2 7B, Context Length=40962024.066.87
INT4-gs128Bits Per FPN=4.16, Backbone=LLama-2-7b2024.056.87
RUQuant#Bit=W4A4, Backbone=LLaMA2-13B2026.046.87
RUQuant#Bit=W6A6, Backbone=LLaMA2-7B2026.046.88
DuQuant#Bit=W4A4, Backbone=LLaMA2-13B2026.046.89
DuQuant#Bit=W6A6, Backbone=LLaMA2-7B2026.046.89
aespaModel Family=LLaMA2, Model Scale=13B, Precision=INT42024.026.891