Normalized Accuracy on ARC-E (Question Answering)
81.1Normalized Accuracy (ARC-E)Liger-GLA
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Liger-GLABase Model=Llama-3-8B, Healing Tokens (B)=0.022026.04 | 81.1 | — | |
| Mistral-7BBase Model=Mistral-7B2026.04 | 80.7 | — | |
| LoLCATsBase Model=Llama-3-8B, Healing Tokens (B)=0.042026.04 | 80.4 | — | |
| Llama-3-8BBase Model=Llama-3-8B2026.04 | 80.1 | — | |
| BaseModel=Mistral-7B2026.04 | 79.5 | — | |
| BaseModel=Llama-70B2026.04 | 79.5 | — | |
| Liger-GLABase Model=Mistral-7B, Healing Tokens (B)=0.022026.04 | 78.7 | — | |
| Qwen3-4BBase Model=Qwen3-4B2026.04 | 78.6 | — | |
| LoLCATsBase Model=Mistral-7B, Healing Tokens (B)=0.042026.04 | 78.4 | — | |
| LayerBoostBase Model=Qwen3-4B, Healing Tokens (B)=0.042026.04 | 77.9 | — | |
| BaseModel=Qwen2.5-7B2026.04 | 77.4 | — | |
| BaseModel=Llama-13B2026.04 | 76.4 | — | |
| SUPRABase Model=Mistral-7B, Healing Tokens (B)=1002026.04 | 75.9 | — | |
| SUPRABase Model=Llama-3-8B, Healing Tokens (B)=202026.04 | 75.1 | — | |
| Mamba2-LlamaBase Model=Llama-3-8B, Healing Tokens (B)=202026.04 | 74.1 | — | |
| LLaMA3Size=3.0B, Bit=16.0, Zero-shot=true2025.08 | 71.6 | — | |
| BinaryLLMSize=3.0B, Bit=1.01, Zero-shot=true2025.08 | 63.1 | — | |
| Nemotron-2 Approx.Shot=0-shot2026.05 | 61.9 | 65.9 | |
| Composer (2Mb-M-3A)Shot=0-shot2026.05 | 61.6 | 65 | |
| AIRAhybrid-AShot=0-shot, Architecture Variant=Stretched2026.05 | 61.5 | 66.6 | |
| AIRAhybrid-DShot=0-shot, Architecture Variant=Stretched2026.05 | 61.4 | 66.3 | |
| BitNetSize=3.0B, Bit=1.59, Zero-shot=true2025.08 | 61.4 | — | |
| Mamba (Mb + M)Shot=0-shot2026.05 | 61.2 | 64.9 | |
| AIRAhybrid-BShot=0-shot, Architecture Variant=Stacked2026.05 | 61.2 | 64.8 | |
| AIRAhybrid-BShot=0-shot, Architecture Variant=Stretched2026.05 | 61.2 | 65.1 | |
| AIRAhybrid-EShot=0-shot, Architecture Variant=Stretched2026.05 | 60.4 | 64.9 | |
| LLaMA3Size=1.3B, Bit=16.0, Zero-shot=true2025.08 | 60.4 | — | |
| AIRAhybrid-DShot=0-shot, Architecture Variant=Stacked2026.05 | 60 | 65.1 | |
| Nemotron-H Approx.Shot=0-shot2026.05 | 59.9 | 65.3 | |
| AIRAhybrid-EShot=0-shot, Architecture Variant=Stacked2026.05 | 59.5 | 64.3 | |
| PythiaSize=2.8B, Bit=16.0, Zero-shot=true2025.08 | 58.8 | — | |
| AIRAhybrid-CShot=0-shot, Architecture Variant=Stretched2026.05 | 58.5 | 65 | |
| SmolLMSize=135M, Bit=16.0, Zero-shot=true2025.08 | 56.3 | — | |
| BitNetSize=1.3B, Bit=1.59, Zero-shot=true2025.08 | 54.9 | — | |
| OPTSize=2.7B, Bit=16.0, Zero-shot=true2025.08 | 54.4 | — | |
| OPTSize=1.3B, Bit=16.0, Zero-shot=true2025.08 | 51 | — | |
| BinaryLLMSize=1.3B, Bit=1.01, Zero-shot=true2025.08 | 47.4 | — | |
| BaseModel=GPT-2 (774M)2026.04 | 46.6 | — | |
| FBI-LLMSize=1.3B, Bit=1.01, Zero-shot=true2025.08 | 43.6 | — | |
| OneBit-OPTSize=2.7B, Bit=1.02, Zero-shot=true2025.08 | 43.4 | — | |
| GPT2-SSource=OpenAI2026.04 | 42.8 | — | |
| BinaryLLMSize=135M, Bit=1.01, Zero-shot=true2025.08 | 41.5 | — | |
| OneBit-OPTSize=1.3B, Bit=1.02, Zero-shot=true2025.08 | 41.3 | — | |
| OPTSize=125M, Bit=16.0, Zero-shot=true2025.08 | 40 | — | |
| PythiaSize=160M, Bit=16.0, Zero-shot=true2025.08 | 39.1 | — | |
| FBI-LLMSize=130M, Bit=1.01, Zero-shot=true2025.08 | 34.9 | — | |
| GPT2-SArchitecture=12MHA2026.04 | 29.4 | — | |
| GPPoM2-S Hyb.Architecture=Hybrid2026.04 | 29 | — | |
| GPPoM2-SArchitecture=Standard2026.04 | 28.7 | — | |
| GAINModel=Qwen2.5-7B2026.04 | 3.8 | — | |
| GAINModel=Mistral-7B2026.04 | 1.5 | — | |
| LoRAModel=Llama-13B2026.04 | 1.5 | — | |
| GAINModel=Llama-70B2026.04 | 1.4 | — | |
| GAINModel=Llama-13B2026.04 | 0.5 | — | |
| LoRAModel=Llama-70B2026.04 | 0.2 | — | |
| LoRAModel=Qwen2.5-7B2026.04 | -0.2 | — | |
| GAINModel=GPT-2 (774M)2026.04 | -1.4 | — | |
| LoRAModel=GPT-2 (774M)2026.04 | -1.9 | — | |
| LoRAModel=Mistral-7B2026.04 | -10.7 | — |