List your company
ResearchFormerly WizWand

Language Modeling on C4 (PPL)

5.52PerplexityFP16
5.41246.13876.8657.5913Oct 16, 2024Jan 28, 2025May 12, 2025Aug 24, 2025Dec 6, 2025Mar 20, 2026Jul 3, 2026
Updated 3mo ago

Evaluation results

MethodLinks
FP16Backbone=Llama-2-70B, Context Length=20482026.065.52
FP16Avg. Bit=16, Backbone=LLaMA2-70B2024.105.52
CMPQAvg. Bit=4, Backbone=LLaMA2-70B2024.105.57
AWQAvg. Bit=4, Backbone=LLaMA2-70B2024.105.58
VPTQ#Backbone=Llama-2-70B, Type=VQ, Bits=3.01-3.03, Context Length=20482026.065.67
QuIP#Backbone=Llama-2-70B, Type=VQ, Bits=3.00, Context Length=20482026.065.67
LiftQuantBackbone=Llama-2-70B, Type=LQ-24/8, Bits=3.01-3.02, Context Length=20482026.065.67
GPTQAvg. Bit=4, Backbone=LLaMA2-70B2024.105.7
EQAT-g128Backbone=Llama-2-70B, Type=UQ, Bits=3.13, Context Length=20482026.065.71
FP16Backbone=Llama-3-70B, Context Length=81922026.065.78
CMPQAvg. Bit=3, Backbone=LLaMA2-70B2024.105.78
RTNAvg. Bit=4, Backbone=LLaMA2-70B2024.105.8
AWQAvg. Bit=3, Backbone=LLaMA2-70B2024.105.81
LiftQuantBackbone=Llama-2-70B, Type=LQ-24/10, Bits=2.41-2.42, Context Length=20482026.065.83
GPTQ-g128Backbone=Llama-2-70B, Type=UQ, Bits=3.13, Context Length=20482026.065.85
QuIPAvg. Bit=4, Backbone=LLaMA2-70B2024.105.86
QTIPBackbone=Llama-2-70B, Type=VQ, Bits=2.00, Context Length=20482026.065.93
LiftQuantBackbone=Llama-2-70B, Type=LQ-32/16, Bits=2.01-2.02, Context Length=20482026.066.06
QuIP#Backbone=Llama-2-70B, Type=VQ, Bits=2.00, Context Length=20482026.066.12
AQLMBackbone=Llama-2-70B, Type=VQ, Bits=1.97-2.07, Context Length=20482026.066.13
VPTQBackbone=Llama-2-70B, Type=VQ, Bits=2.02-2.08, Context Length=20482026.066.13
QuIPAvg. Bit=3, Backbone=LLaMA2-70B2024.106.14
QuIP#Backbone=Llama-3-70B, Type=VQ, Bits=3.00, Context Length=81922026.066.18
LiftQuantBackbone=Llama-3-70B, Type=LQ-24/8, Bits=3.01-3.02, Context Length=81922026.066.18
EQAT-g128Backbone=Llama-3-70B, Type=UQ, Bits=3.13, Context Length=81922026.066.3
EQAT-g64Backbone=Llama-2-70B, Type=UQ, Bits=2.25, Context Length=20482026.066.38
BF16Algorithm=BF16, Bits=16, Backbone=Llama 3.1 70B Instruct2025.056.46
BF16Base Model=Llama 3.1 70B Instruct, Bits=162025.056.46
FP16Backbone=Llama-2-13B, Context Length=20482026.066.47
LiftQuantBackbone=Llama-3-70B, Type=LQ-24/10, Bits=2.41-2.42, Context Length=81922026.066.47
FP16Backbone=2-132026.066.47
YAQA-BAlgorithm=YAQA-B, Bits=4, Backbone=Llama 3.1 70B Instruct2025.056.51
YAQA-ABase Model=Llama 3.1 70B Instruct, Bits=4, Quantizer=QTIP, Recovery Finetuning=true2025.056.52
YAQA-BBase Model=Llama 3.1 70B Instruct, Bits=4, Quantizer=QTIP, Recovery Finetuning=true2025.056.52
LDLQBase Model=Llama 3.1 70B Instruct, Bits=4, Quantizer=QTIP, Recovery Finetuning=true2025.056.54
GPTQBackbone=Llama-2-70B, Type=UQ, Bits=3.00, Context Length=20482026.066.57
YAQA-BBase Model=Llama 3.1 70B Instruct, Bits=3, Quantizer=QTIP, Recovery Finetuning=true2025.056.61
YAQA-ABase Model=Llama 3.1 70B Instruct, Bits=3, Quantizer=QTIP, Recovery Finetuning=true2025.056.63
LDLQBase Model=Llama 3.1 70B Instruct, Bits=3, Quantizer=QTIP, Recovery Finetuning=true2025.056.64
GPTQAvg. Bit=3, Backbone=LLaMA2-70B2024.106.69
VPTQ#Backbone=Llama-2-13B, Type=VQ, Bits=3.01-3.03, Context Length=20482026.066.7
LiftQuantBackbone=Llama-2-13B, Type=LQ-24/8, Bits=3.01-3.02, Context Length=20482026.066.71
QuIP#Backbone=Llama-2-13B, Type=VQ, Bits=3.00, Context Length=20482026.066.72
LiftQuantBackbone=Llama-3-70B, Type=LQ-32/16, Bits=2.01-2.02, Context Length=81922026.066.73
EQAT-g128Backbone=Llama-2-13B, Type=UQ, Bits=3.13, Context Length=20482026.066.73
QTIPBackbone=Llama-3-70B, Type=VQ, Bits=2.00, Context Length=81922026.066.8
FP16Backbone=Llama-2-7B, Context Length=20482026.066.97
FP16Backbone=2-72026.066.97
LiftQuantBackbone=Llama-2-13B, Type=LQ-24/10, Bits=2.41-2.42, Context Length=20482026.066.98
GPTQ-g128Backbone=Llama-2-13B, Type=UQ, Bits=3.13, Context Length=20482026.067
YAQA-BBase Model=Llama 3.1 70B Instruct, Bits=2, Quantizer=QTIP, Recovery Finetuning=true2025.057.07
FP16Backbone=Llama-3-8B, Context Length=81922026.067.1
YAQA-ABase Model=Llama 3.1 70B Instruct, Bits=2, Quantizer=QTIP, Recovery Finetuning=true2025.057.1
QTIPBackbone=Llama-2-13B, Type=VQ, Bits=2.00, Context Length=20482026.067.16
LDLQBase Model=Llama 3.1 70B Instruct, Bits=2, Quantizer=QTIP, Recovery Finetuning=true2025.057.16
VPTQBackbone=Llama-3-70B, Type=VQ, Bits=2.02-2.08, Context Length=81922026.067.3
LiftQuantBackbone=Llama-2-7B, Type=LQ-24/8, Bits=3.01-3.02, Context Length=20482026.067.31
QuIP#Backbone=Llama-2-7B, Type=VQ, Bits=3.00, Context Length=20482026.067.32
VPTQ#Backbone=Llama-2-7B, Type=VQ, Bits=3.01-3.03, Context Length=20482026.067.33
OriginalCompression Ratio=0%2024.117.34
LiftQuantBackbone=Llama-2-13B, Type=LQ-32/16, Bits=2.01-2.02, Context Length=20482026.067.34
EQAT-g128Backbone=Llama-2-7B, Type=UQ, Bits=3.13, Context Length=20482026.067.34
YAQA-BAlgorithm=YAQA-B, Bits=2, Backbone=Llama 3.1 70B Instruct2025.057.34
Base ModelParameter Rate=100%2026.067.34
BaselineRatio=0.0, Backbone=LLaMA-7B, Fine-tuning=false2026.077.34
GPTQ-g128Backbone=Llama-3-70B, Type=UQ, Bits=3.13, Context Length=81922026.067.37
VPTQBackbone=Llama-2-13B, Type=VQ, Bits=2.02-2.08, Context Length=20482026.067.41
EQAT-g64Backbone=Llama-3-70B, Type=UQ, Bits=2.25, Context Length=81922026.067.43
AQLMBackbone=Llama-2-13B, Type=VQ, Bits=1.97-2.07, Context Length=20482026.067.44
QuIP#Backbone=Llama-2-13B, Type=VQ, Bits=2.00, Context Length=20482026.067.45
QuIP#Backbone=Llama-3-70B, Type=VQ, Bits=2.00, Context Length=81922026.067.46
NeUQISetting=W4A4, Backbone=LLaMA 2 7B, Hadamard Transform=true2025.057.57
DbyD (35W)Model=Mistral 7B, Codebooks=5 × 82026.067.58
MagRSetting=W4A4, Backbone=LLaMA 2 7B, Hadamard Transform=true2025.057.59
EQAT-g64Backbone=Llama-2-13B, Type=UQ, Bits=2.25, Context Length=20482026.067.59
OmniQ-g64Backbone=Llama-2-70B, Type=UQ, Bits=2.25, Context Length=20482026.067.68
LiftQuantBackbone=Llama-2-7B, Type=LQ-24/10, Bits=2.41-2.42, Context Length=20482026.067.7
LiftQuantBackbone=Llama-3-8B, Type=LQ-24/8, Bits=3.01-3.02, Context Length=81922026.067.7
DbyD (Uni.)Model=Mistral 7B, Codebooks=5 × 82026.067.7
QuIP#Backbone=Llama-3-8B, Type=VQ, Bits=3.00, Context Length=81922026.067.71
GPTAQSetting=W4A4, Backbone=LLaMA 2 7B, Hadamard Transform=true2025.057.73
AQLM (Ind.)Model=Mistral 7B, Codebooks=4 × 82026.067.73
GPTQSetting=W4A4, Backbone=LLaMA 2 7B, Hadamard Transform=true2025.057.75
AQLM (Drop)Model=MetaLlama 8B, Codebooks=3 × 82026.067.78
EQAT-g128Backbone=Llama-3-8B, Type=UQ, Bits=3.13, Context Length=81922026.067.79
GPTQ-g128Backbone=Llama-2-7B, Type=UQ, Bits=3.13, Context Length=20482026.067.89
AQLM (Ind.)Model=Mistral 7B, Codebooks=5 × 82026.067.93
AQLM (Drop)Model=Mistral 7B, Codebooks=5 × 82026.067.93
QTIPBackbone=Llama-2-7B, Type=VQ, Bits=2.00, Context Length=20482026.067.94
GPTQBackbone=Llama-2-13B, Type=UQ, Bits=3.00, Context Length=20482026.068.02
BF16Algorithm=BF16, Bits=16, Backbone=Llama 3.1 8B Instruct2025.058.02
BF16Bits=16, Finetuning=none, Incoherence processing=true, Backbone=Llama 3.1 8B Instruct2025.058.02
BF16Base Model=Llama 3.1 8B Instruct, Bits=162025.058.02
DbyD (35W)Model=Mistral 7B, Codebooks=4 × 82026.068.08
FiPSCompression Ratio=20%2024.118.1
YAQA-ABase Model=Llama 3.1 8B Instruct, Bits=4, Quantizer=QTIP, Recovery Finetuning=true2025.058.1
YAQA-BBase Model=Llama 3.1 8B Instruct, Bits=4, Quantizer=QTIP, Recovery Finetuning=true2025.058.11
YAQA-BAlgorithm=YAQA-B, Bits=4, Backbone=Llama 3.1 8B Instruct2025.058.12
LDLQBase Model=Llama 3.1 8B Instruct, Bits=4, Quantizer=QTIP, Recovery Finetuning=true2025.058.13
LiftQuantBackbone=Llama-2-7B, Type=LQ-32/16, Bits=2.01-2.02, Context Length=20482026.068.21