List your company
ResearchFormerly WizWand

Language Modeling on WikiText-2 (DKL and PPL)

3.52PPLBF16
1.549614.849828.1541.4502May 29, 2025Aug 1, 2025Oct 5, 2025Dec 8, 2025Feb 11, 2026Apr 16, 2026Jun 20, 2026
Updated 4mo ago

Evaluation results

MethodLinks
BF16Algorithm=BF16, Bits=16, Backbone=Llama 3.1 70B Instruct2025.053.520
BF16Base Model=Llama 3.1 70B Instruct, Bits=162025.053.520
YAQA-BBase Model=Llama 3.1 70B Instruct, Bits=4, Quantizer=QTIP, Recovery Finetuning=true, Incoherence processing=true2025.053.670.029
YAQA-ABase Model=Llama 3.1 70B Instruct, Bits=4, Quantizer=QTIP, Recovery Finetuning=true, Incoherence processing=true2025.053.680.032
YAQA-BAlgorithm=YAQA-B, Bits=4, Backbone=Llama 3.1 70B Instruct2025.053.690.03
LDLQBase Model=Llama 3.1 70B Instruct, Bits=4, Quantizer=QTIP, Recovery Finetuning=true, Incoherence processing=true2025.053.710.036
GuidedQuantAlgorithm=GuidedQuant, Bits=4, Backbone=Llama 3.1 70B Instruct2025.053.720.035
YAQA-AAlgorithm=YAQA-A, Bits=4, Backbone=Llama 3.1 70B Instruct2025.053.730.036
LDLQAlgorithm=LDLQ, Bits=4, Backbone=Llama 3.1 70B Instruct2025.053.740.045
YAQA-BBase Model=Llama 3.1 70B Instruct, Bits=3, Quantizer=QTIP, Recovery Finetuning=true, Incoherence processing=true2025.053.870.091
YAQA-ABase Model=Llama 3.1 70B Instruct, Bits=3, Quantizer=QTIP, Recovery Finetuning=true, Incoherence processing=true2025.053.880.098
LDLQBase Model=Llama 3.1 70B Instruct, Bits=3, Quantizer=QTIP, Recovery Finetuning=true, Incoherence processing=true2025.053.960.101
YAQA-BAlgorithm=YAQA-B, Bits=3, Backbone=Llama 3.1 70B Instruct2025.054.010.094
GuidedQuantAlgorithm=GuidedQuant, Bits=3, Backbone=Llama 3.1 70B Instruct2025.054.10.111
YAQA-AAlgorithm=YAQA-A, Bits=3, Backbone=Llama 3.1 70B Instruct2025.054.10.11
LDLQAlgorithm=LDLQ, Bits=3, Backbone=Llama 3.1 70B Instruct2025.054.260.138
YAQA-BBase Model=Llama 3.1 70B Instruct, Bits=2, Quantizer=QTIP, Recovery Finetuning=true, Incoherence processing=true2025.054.820.266
YAQA-ABase Model=Llama 3.1 70B Instruct, Bits=2, Quantizer=QTIP, Recovery Finetuning=true, Incoherence processing=true2025.054.920.279
LDLQBase Model=Llama 3.1 70B Instruct, Bits=2, Quantizer=QTIP, Recovery Finetuning=true, Incoherence processing=true2025.055.010.302
YAQA-BAlgorithm=YAQA-B, Bits=2, Backbone=Llama 3.1 70B Instruct2025.055.30.335
YAQA-AAlgorithm=YAQA-A, Bits=2, Backbone=Llama 3.1 70B Instruct2025.055.560.378
GuidedQuantAlgorithm=GuidedQuant, Bits=2, Backbone=Llama 3.1 70B Instruct2025.055.590.383
LDLQAlgorithm=LDLQ, Bits=2, Backbone=Llama 3.1 70B Instruct2025.056.020.497
BF16Algorithm=BF16, Bits=16, Backbone=Llama 3.1 8B Instruct2025.056.50
BF16Base Model=Llama 3.1 8B Instruct, Bits=162025.056.50
YAQA-BBase Model=Llama 3.1 8B Instruct, Bits=4, Quantizer=QTIP, Recovery Finetuning=true, Incoherence processing=true2025.056.560.012
YAQA-ABase Model=Llama 3.1 8B Instruct, Bits=4, Quantizer=QTIP, Recovery Finetuning=true, Incoherence processing=true2025.056.580.014
YAQA-BAlgorithm=YAQA-B, Bits=4, Backbone=Llama 3.1 8B Instruct2025.056.610.013
LDLQBase Model=Llama 3.1 8B Instruct, Bits=4, Quantizer=QTIP, Recovery Finetuning=true, Incoherence processing=true2025.056.610.016
YAQA-BBase Model=Llama 3.1 8B Instruct, Bits=3, Quantizer=QTIP, Recovery Finetuning=true, Incoherence processing=true2025.056.740.038
YAQA-ABase Model=Llama 3.1 8B Instruct, Bits=3, Quantizer=QTIP, Recovery Finetuning=true, Incoherence processing=true2025.056.780.042
LDLQBase Model=Llama 3.1 8B Instruct, Bits=3, Quantizer=QTIP, Recovery Finetuning=true, Incoherence processing=true2025.056.80.048
DenseSparsity=0%, Backbone=Llama2-7B2026.067.33—
GOOGLE QATQuantization Type=QAT, Bits=4.52025.057.560.089
YAQA-BBase Model=Llama 3.1 8B Instruct, Bits=2, Quantizer=QTIP, Recovery Finetuning=true, Incoherence processing=true2025.057.60.147
YAQA-ABase Model=Llama 3.1 8B Instruct, Bits=2, Quantizer=QTIP, Recovery Finetuning=true, Incoherence processing=true2025.057.630.163
LDLQBase Model=Llama 3.1 8B Instruct, Bits=2, Quantizer=QTIP, Recovery Finetuning=true, Incoherence processing=true2025.057.820.185
BF16Quantization Type=NONE, Bits=162025.057.850
YAQA-BQuantization Type=PTQ, Bits=42025.057.940.056
YAQA-AQuantization Type=PTQ, Bits=42025.057.960.058
UNQUANTIZEDBITS=16, Context window=4K2025.0590
YAQA-BBITS=4, Context window=4K2025.059.040.0153
LDLQBITS=4, Context window=4K2025.059.070.0216
YAQA-BBITS=3, Context window=4K2025.059.20.0479
LDLQBITS=3, Context window=4K2025.059.310.0681
BF16Algorithm=BF16, Bits=16, Backbone=Llama 3.2 3B Instruct2025.059.580
YAQA-BAlgorithm=YAQA-B, Bits=4, Backbone=Llama 3.2 3B Instruct2025.059.80.014
YAQA-BBITS=2, Context window=4K2025.0510.160.205
LDLQBITS=2, Context window=4K2025.0510.810.285
BF16Algorithm=BF16, Bits=16, Backbone=Llama 3.2 1B Instruct2025.0511.570
YAQA-BAlgorithm=YAQA-B, Bits=4, Backbone=Llama 3.2 1B Instruct2025.0511.830.019
UniRankSparsity=40%, Backbone=Llama2-7B, Base Method=LoRAP2026.0614.03—
SkipGPTSparsity=40%, Backbone=Llama2-7B2026.0614.35—
SliceGPTSparsity=40%, Backbone=Llama2-7B2026.0614.87—
LoRAPSparsity=40%, Backbone=Llama2-7B2026.0616.08—
DRDG/TWEvidence form=Dual-end output aggregation (token-wise sync)2026.0617.016—
DRAGONEvidence form=Dual-end output aggregation2026.0617.016—
CONCORDEvidence form=Dual-end output aggregation2026.0617.016—
DRCGEvidence form=Dual-end evidence merged into single-end context2026.0617.179—
Shortened-llamaSparsity=40%, Backbone=Llama2-7B2026.0617.22—
ShortGPTSparsity=40%, Backbone=Llama2-7B2026.0618.35—
CRCGEvidence form=Single-end local shard2026.0621.351—
MoDSparsity=40%, Backbone=Llama2-7B2026.0640.42—
D-LLMSparsity=40%, Backbone=Llama2-7B2026.0652.78—