List your company
ResearchFormerly WizWand

Compositional Reasoning on GQA

62Accuracy (GQA Compositional Reasoning)FP16 (Baseline)
1.99217.57133.1548.729Sep 28, 2025
Updated 4mo ago

Evaluation results

MethodLinks
FP16 (Baseline)Backbone=LLaVA-1.5 7B, Avg. Bit width=162025.0962
AWQBackbone=LLaVA-1.5 7B, Avg. Bit width=42025.0961.5
GPTQBackbone=LLaVA-1.5 7B, Avg. Bit width=42025.0961.4
FP16 (Baseline)Backbone=Qwen 2.5 VL 7B, Avg. Bit width=162025.0960.5
AWQBackbone=Qwen 2.5 VL 7B, Avg. Bit width=42025.0960.5
GPTQBackbone=Qwen 2.5 VL 7B, Avg. Bit width=42025.0960.1
GPTQBackbone=LLaVA-1.5 7B, Avg. Bit width=32025.0958.8
LUQBackbone=LLaVA-1.5 7B, Avg. Bit width=2.54, Quantized Layers=16 layer2025.0958.2
OmniQuantBackbone=Qwen 2.5 VL 7B, Avg. Bit width=42025.0957.5
GPTQBackbone=Qwen 2.5 VL 7B, Avg. Bit width=32025.0953.4
LUQBackbone=Qwen 2.5 VL 7B, Avg. Bit width=2.75, Quantized Layers=12 layer2025.0952.9
OmniQuantBackbone=LLaVA-1.5 7B, Avg. Bit width=42025.0939.2
SlimLLMBackbone=LLaVA-1.5 7B, Avg. Bit width=42025.0937.7
SlimLLMBackbone=LLaVA-1.5 7B, Avg. Bit width=32025.0934.4
BiLLMBackbone=LLaVA-1.5 7B, Avg. Bit width=1.082025.0922.7
OmniQuantBackbone=LLaVA-1.5 7B, Avg. Bit width=32025.0915.1
OmniQuantBackbone=Qwen 2.5 VL 7B, Avg. Bit width=32025.0912.8
BiLLMBackbone=Qwen 2.5 VL 7B, Avg. Bit width=1.082025.094.3