List your company
ResearchFormerly WizWand

Text Generation Accuracy on GSM8K

86.4AccuracyLoRA + PCGrad
10.417630.143849.8769.5962May 28, 2026Jun 1, 2026Jun 5, 2026Jun 9, 2026Jun 13, 2026Jun 17, 2026Jun 22, 2026
Updated 4mo ago

Evaluation results

MethodLinks
LoRA + PCGradBackbone=Qwen-14B2026.0686.4
GRAINBackbone=Qwen-14B2026.0685.53
LoRA + SWABackbone=Qwen-14B2026.0685.45
LoRA + SAMBackbone=Qwen-14B2026.0685.45
Llama 3.1 8B Instruct# Bits=BF16, Decoding Strategy=greedy2026.0584.99
Llama 3.1 8B Instruct# Bits=BF16, Decoding strategy=greedy2026.0584.99
LoRA + NoisyTuneBackbone=Qwen-14B2026.0684.46
LoRABackbone=Qwen-14B2026.0682.94
LoRA + FocalLossBackbone=Qwen-14B2026.0682.18
FlexRound+LFQ# Bits=W4, PTQ Method=FlexRound, Decoding Strategy=greedy, Quantization Granularity=4-bit per-channel weight-only2026.0581.8
FlexRound+LFQ# Bits=W4, Decoding strategy=greedy2026.0581.8
FlexRound+LFQ+RILQ# Bits=W4, PTQ Method=FlexRound, Decoding Strategy=greedy, Quantization Granularity=4-bit per-channel weight-only2026.0581.43
FlexRound# Bits=W4, PTQ Method=FlexRound, Decoding Strategy=greedy, Quantization Granularity=4-bit per-channel weight-only2026.0581.35
FlexRound# Bits=W4, Decoding strategy=greedy2026.0581.35
FlexRound+RILQ# Bits=W4, PTQ Method=FlexRound, Decoding Strategy=greedy, Quantization Granularity=4-bit per-channel weight-only2026.0580.52
FlexRound+RILQ# Bits=W4, Decoding strategy=greedy2026.0580.52
GRAINBackbone=Ministral-14B2026.0679.85
OmniQuant+LFQ# Bits=W4, PTQ Method=OmniQuant, Decoding Strategy=greedy, Quantization Granularity=4-bit per-channel weight-only2026.0579.76
OmniQuant+LFQ# Bits=W4, Decoding strategy=greedy2026.0579.76
LoRA + SAMBackbone=Ministral-14B2026.0679.63
OmniQuant+LFQ+RILQ# Bits=W4, PTQ Method=OmniQuant, Decoding Strategy=greedy, Quantization Granularity=4-bit per-channel weight-only2026.0579.45
LoRABackbone=Ministral-14B2026.0679.24
LoRA + PCGradBackbone=Ministral-14B2026.0679.12
OmniQuant+RILQ# Bits=W4, PTQ Method=OmniQuant, Decoding Strategy=greedy, Quantization Granularity=4-bit per-channel weight-only2026.0578.85
OmniQuant+RILQ# Bits=W4, Decoding strategy=greedy2026.0578.85
LoRA + NoisyTuneBackbone=Ministral-14B2026.0678.77
LoRA + SWABackbone=Ministral-14B2026.0678.77
OmniQuant# Bits=W4, PTQ Method=OmniQuant, Decoding Strategy=greedy, Quantization Granularity=4-bit per-channel weight-only2026.0578.17
OmniQuant# Bits=W4, Decoding strategy=greedy2026.0578.17
LoRA + FocalLossBackbone=Ministral-14B2026.0676.42
GRAINBackbone=Qwen-7B2026.0674.77
LoRA + PCGradBackbone=Qwen-7B2026.0674.63
Block-AP+LFQ# Bits=W4, PTQ Method=Block-AP, Decoding Strategy=greedy, Quantization Granularity=4-bit per-channel weight-only2026.0574.45
Block-AP+LFQ# Bits=W4, Decoding strategy=greedy2026.0574.45
LoRA + NoisyTuneBackbone=Qwen-7B2026.0674.43
Block-AP+LFQ+RILQ# Bits=W4, PTQ Method=Block-AP, Decoding Strategy=greedy, Quantization Granularity=4-bit per-channel weight-only2026.0574.22
LoRA + FocalLossBackbone=Qwen-7B2026.0673.98
Block-AP+RILQ# Bits=W4, PTQ Method=Block-AP, Decoding Strategy=greedy, Quantization Granularity=4-bit per-channel weight-only2026.0573.92
Block-AP+RILQ# Bits=W4, Decoding strategy=greedy2026.0573.92
LoRA + SWABackbone=Qwen-7B2026.0673.9
Block-AP# Bits=W4, PTQ Method=Block-AP, Decoding Strategy=greedy, Quantization Granularity=4-bit per-channel weight-only2026.0573.84
Block-AP# Bits=W4, Decoding strategy=greedy2026.0573.84
LoRA + SAMBackbone=Qwen-7B2026.0672.88
LoRABackbone=Qwen-7B2026.0672.87
DeLaskModel=LLaMA3-8B, Decoding=DeLask2026.0555.42
UADModel=LLaMA3-8B, Decoding=UAD2026.0555.02
RegularModel=LLaMA3-8B, Decoding=Regular2026.0554.89
DOLAModel=LLaMA3-8B, Decoding=DOLA2026.0554.74
ITIModel=LLaMA3-8B, Decoding=ITI2026.0554.36
ADModel=LLaMA3-8B, Decoding=AD2026.0553.39
GRAINBackbone=Mistral-7B2026.0649.67
LoRA + SWABackbone=Mistral-7B2026.0648.65
LoRA + NoisyTuneBackbone=Mistral-7B2026.0648.2
LoRA + PCGradBackbone=Mistral-7B2026.0648.14
LoRABackbone=Mistral-7B2026.0647.95
LoRA + FocalLossBackbone=Mistral-7B2026.0647.53
LoRA + SAMBackbone=Mistral-7B2026.0646.49
DeLaskModel=LLaMA2-7B, Decoding=DeLask2026.0515.04
UADModel=LLaMA2-7B, Decoding=UAD2026.0514.86
ITIModel=LLaMA2-7B, Decoding=ITI2026.0514.74
RegularModel=LLaMA2-7B, Decoding=Regular2026.0514.7
ADModel=LLaMA2-7B, Decoding=AD2026.0513.76
DOLAModel=LLaMA2-7B, Decoding=DOLA2026.0513.34