List your company
ResearchFormerly WizWand

Zero-shot Evaluation on Downstream Tasks

76AccuracyFP16
60.81664.75868.772.642May 16, 2025Jun 18, 2025Jul 22, 2025Aug 24, 2025Sep 27, 2025Oct 30, 2025Dec 3, 2025
Updated 8mo ago

Evaluation results

MethodLinks
FP16Weight bit=16, Pruning Ratio=0%, Reduction size=0x, Backbone=Nemotron-H 8B2025.1276
UniQLone-pass=true, Fine-tuning=true, Weight bit=4, Pruning Ratio=0%, Reduction size=4x, Backbone=Bamba-v2 9B2025.1275.1
FP16Weight bit=16, Pruning Ratio=0%, Reduction size=0x, Backbone=Bamba-v2 9B2025.1274.6
BF16Model Size=8B, Precision=BF162025.0574.4
BF16Model Size=8B2025.0574.4
BF16Model Size=8B2025.0574.1
FP16Weight bit=16, Pruning Ratio=0%, Reduction size=0x, Backbone=Llama-3.1 8B2025.1274
UniQLone-pass=true, Fine-tuning=true, Weight bit=4, Pruning Ratio=0%, Reduction size=4x, Backbone=Llama-3.1 8B2025.1273.6
BF16Stage 1=None, Model Size=8B, Quantization Bits=None2025.0573.6
UniQLone-pass=true, Fine-tuning=true, Weight bit=4, Pruning Ratio=0%, Reduction size=4x, Backbone=Nemotron-H 8B2025.1273.3
GPTAQStage 1=HIP, Bits=W4, Model Size=8B2025.0572.7
FP16Weight bit=16, Pruning Ratio=0%, Reduction size=0x, Backbone=Qwen-2.5 7B2025.1272.4
UniQLone-pass=true, Fine-tuning=true, Weight bit=4, Pruning Ratio=0%, Reduction size=4x, Backbone=Qwen-2.5 7B2025.1272.4
QronosStage 1=None, Bits=W4, Model Size=8B2025.0572.2
OPTQStage 1=HIP, Bits=W4, Model Size=8B2025.0572.1
QronosStage 1=HIP, Bits=W4, Model Size=8B2025.0572.1
GPFQStage 1=HIP, Bits=W4, Model Size=8B2025.0571.9
OPTQStage 1=None, Bits=W4, Model Size=8B2025.0571.8
GPFQStage 1=None, Bits=W4, Model Size=8B2025.0571.5
UniQLone-pass=true, Fine-tuning=true, Weight bit=4, Pruning Ratio=15%, Reduction size=4.7x, Backbone=Llama-3.1 8B2025.1271.4
GPTAQStage 1=None, Bits=W4, Model Size=8B2025.0571.2
RTNStage 1=HIP, Bits=W4, Model Size=8B2025.0570.8
FP16Weight bit=16, Pruning Ratio=0%, Reduction size=0x, Backbone=Mamba2 8B2025.1270.6
UniQLone-pass=true, Fine-tuning=true, Weight bit=4, Pruning Ratio=15%, Reduction size=4.7x, Backbone=Nemotron-H 8B2025.1270.5
UniQLone-pass=true, Fine-tuning=true, Weight bit=4, Pruning Ratio=15%, Reduction size=4.7x, Backbone=Bamba-v2 9B2025.1270.3
BF16Stage 1=None, Model Size=4B, Quantization Bits=None2025.0570.1
QronosTransformation=SmoothRot, Precision=W4A4KV16, Model Size=8B2025.0569.4
UniQLone-pass=true, Fine-tuning=true, Weight bit=4, Pruning Ratio=0%, Reduction size=4x, Backbone=Mamba2 8B2025.1269.3
QronosTransformation=QuaRot, Precision=W4A4KV16, Model Size=8B2025.0568.9
FP16Weight bit=16, Pruning Ratio=0%, Reduction size=0x, Backbone=Llama-2 7B2025.1268.8
GPTAQTransformation=SmoothRot, Precision=W4A4KV16, Model Size=8B2025.0568.8
OPTQTransformation=SmoothRot, Precision=W4A4KV16, Model Size=8B2025.0568.5
QronosTransformation=SpinQuant, Precision=W4A4KV16, Model Size=8B2025.0568.4
GPTAQTransformation=SpinQuant, Precision=W4A4KV16, Model Size=8B2025.0568.3
UniQLone-pass=true, Fine-tuning=true, Weight bit=4, Pruning Ratio=15%, Reduction size=4.7x, Backbone=Qwen-2.5 7B2025.1268.1
GPTAQTransformation=QuaRot, Precision=W4A4KV16, Model Size=8B2025.0568.1
QronosStage 1=HIP, Stage 2=Qronos, Model Size=8B, Quantization Bits=W3, Beta=0.92025.0568
UniQLone-pass=true, Fine-tuning=true, Weight bit=4, Pruning Ratio=25%, Reduction size=5.3x, Backbone=Llama-3.1 8B2025.1267.7
GPFQTransformation=SmoothRot, Precision=W4A4KV16, Model Size=8B2025.0567.7
UniQLone-pass=true, Fine-tuning=true, Weight bit=4, Pruning Ratio=0%, Reduction size=4x, Backbone=Llama-2 7B2025.1267.6
RTNStage 1=None, Bits=W4, Model Size=8B2025.0567.6
OPTQStage 1=HIP, Stage 2=OPTQ, Model Size=8B, Quantization Bits=W3, Beta=0.92025.0567.6
GPFQTransformation=QuaRot, Precision=W4A4KV16, Model Size=8B2025.0567.6
BF16Model Size=3B, Precision=BF162025.0567.5
BF16Model Size=3B2025.0567.5
UniQLone-pass=true, Fine-tuning=true, Weight bit=4, Pruning Ratio=25%, Reduction size=5.3x, Backbone=Bamba-v2 9B2025.1267.4
OPTQTransformation=SpinQuant, Precision=W4A4KV16, Model Size=8B2025.0567
GPFQTransformation=SpinQuant, Precision=W4A4KV16, Model Size=8B2025.0567
SVD-LLMone-pass=false, Fine-tuning=true, Weight bit=4 (FP16 embeddings/output), Pruning Ratio=15%, Reduction size=4.7x, Backbone=Qwen-2.5 7B2025.1266.8
OPTQTransformation=QuaRot, Precision=W4A4KV16, Model Size=8B2025.0566.7
BF16Model Size=3B2025.0566.4
GPTAQStage 1=HIP, Stage 2=GPTAQ, Model Size=8B, Quantization Bits=W3, Beta=0.92025.0566
UniQLone-pass=true, Fine-tuning=true, Weight bit=4, Pruning Ratio=15%, Reduction size=4.7x, Backbone=Mamba2 8B2025.1265.8
QronosTransformation=SpinQuant, Precision=W4A4KV4, Model Size=8B2025.0565.8
UniQLone-pass=true, Fine-tuning=true, Weight bit=4, Pruning Ratio=15%, Reduction size=4.7x, Backbone=Llama-2 7B2025.1265.6
OPTQTransformation=SpinQuant, Precision=W4A4KV4, Model Size=8B2025.0565.5
GPFQStage 1=HIP, Bits=W4, Model Size=3B2025.0565.3
GPTAQTransformation=SpinQuant, Precision=W4A4KV4, Model Size=8B2025.0565.3
QronosTransformation=SmoothRot, Precision=W4A4KV4, Model Size=8B2025.0565.2
QronosStage 1=None, Bits=W4, Model Size=3B2025.0564.8
QronosStage 1=HIP, Bits=W4, Model Size=3B2025.0564.8
GPTAQTransformation=QuaRot, Precision=W4A4KV4, Model Size=8B2025.0564.8
QronosTransformation=QuaRot, Precision=W4A4KV4, Model Size=8B2025.0564.8
UniQLone-pass=true, Fine-tuning=true, Weight bit=4, Pruning Ratio=25%, Reduction size=5.3x, Backbone=Nemotron-H 8B2025.1264.7
SVD-LLMone-pass=false, Fine-tuning=true, Weight bit=4 (FP16 embeddings/output), Pruning Ratio=25%, Reduction size=4.7x, Backbone=Qwen-2.5 7B2025.1264.6
OPTQStage 1=HIP, Bits=W4, Model Size=3B2025.0564.5
GPFQStage 1=None, Bits=W4, Model Size=3B2025.0564.4
OPTQStage 1=HIP, Stage 2=OPTQ, Model Size=4B, Quantization Bits=W3, Beta=0.92025.0564.4
MoDeGPTone-pass=false, Fine-tuning=true, Weight bit=4, Pruning Ratio=15%, Reduction size=4.7x, Backbone=Llama-3.1 8B2025.1264.2
RTNTransformation=SpinQuant, Precision=W4A4KV16, Model Size=8B2025.0564.2
QronosStage 1=HIP, Bits=W3, Model Size=8B2025.0564.1
OPTQTransformation=QuaRot, Precision=W4A4KV4, Model Size=8B2025.0564.1
UniQLone-pass=true, Fine-tuning=true, Weight bit=4, Pruning Ratio=25%, Reduction size=5.3x, Backbone=Qwen-2.5 7B2025.1264
GPTAQStage 1=HIP, Bits=W4, Model Size=3B2025.0563.9
BF16Stage 1=None, Model Size=1.7B, Quantization Bits=None2025.0563.9
MoDeGPTone-pass=false, Fine-tuning=true, Weight bit=4, Pruning Ratio=15%, Reduction size=4.7x, Backbone=Llama-2 7B2025.1263.7
GPTAQStage 1=None, Stage 2=GPTAQ, Model Size=4B, Quantization Bits=W3, Beta=0.92025.0563.7
GPTAQTransformation=SmoothRot, Precision=W4A4KV4, Model Size=8B2025.0563.7
OPTQStage 1=None, Bits=W4, Model Size=3B2025.0563.6
RTNStage 1=HIP, Bits=W4, Model Size=3B2025.0563.6
GPTAQStage 1=HIP, Stage 2=GPTAQ, Model Size=4B, Quantization Bits=W3, Beta=0.92025.0563.6
UniQLone-pass=true, Fine-tuning=true, Weight bit=4, Pruning Ratio=25%, Reduction size=5.3x, Backbone=Llama-2 7B2025.1263.5
QronosStage 1=HIP, Stage 2=Qronos, Model Size=4B, Quantization Bits=W3, Beta=0.92025.0563.5
SVD-LLMone-pass=false, Fine-tuning=true, Weight bit=4 (FP16 embeddings/output), Pruning Ratio=15%, Reduction size=4.7x, Backbone=Llama-2 7B2025.1263.2
GPTAQStage 1=HIP, Bits=W3, Model Size=8B2025.0563.2
GPTAQStage 1=None, Bits=W4, Model Size=3B2025.0563.1
UniQLone-pass=true, Fine-tuning=true, Weight bit=4, Pruning Ratio=35%, Reduction size=6.1x, Backbone=Llama-3.1 8B2025.1262.7
UniQLone-pass=true, Fine-tuning=true, Weight bit=4, Pruning Ratio=35%, Reduction size=6.1x, Backbone=Bamba-v2 9B2025.1262.7
RTNTransformation=QuaRot, Precision=W4A4KV16, Model Size=8B2025.0562.6
RTNTransformation=SmoothRot, Precision=W4A4KV16, Model Size=8B2025.0562.6
RTNStage 1=None, Bits=W4, Model Size=3B2025.0562.3
GPFQStage 1=HIP, Stage 2=GPFQ, Model Size=8B, Quantization Bits=W3, Beta=0.92025.0562.2
QronosTransformation=SpinQuant, Precision=W4A4KV16, Model Size=3B2025.0562.1
GPFQTransformation=SmoothRot, Precision=W4A4KV16, Model Size=3B2025.0562
UniQLone-pass=true, Fine-tuning=true, Weight bit=4, Pruning Ratio=25%, Reduction size=5.3x, Backbone=Mamba2 8B2025.1261.8
OPTQStage 1=None, Stage 2=OPTQ, Model Size=8B, Quantization Bits=W3, Beta=0.92025.0561.7
GPFQStage 1=HIP, Stage 2=GPFQ, Model Size=4B, Quantization Bits=W3, Beta=0.92025.0561.6
QronosStage 1=None, Stage 2=Qronos, Model Size=8B, Quantization Bits=W3, Beta=0.92025.0561.5
QronosTransformation=QuaRot, Precision=W4A4KV16, Model Size=3B2025.0561.5
GPTAQTransformation=QuaRot, Precision=W4A4KV16, Model Size=3B2025.0561.4