Conversational Ability on MT-Bench (Score and Avg. Time)
7.58MT-Bench ScoreQwen3-32B
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Qwen3-32BSize=32B2026.03 | 7.58 | — | — | — | |
| Gemma-3-27B-itSize=27B2026.03 | 7.24 | — | — | — | |
| Llama-3.3-70b-InstructSize=70B2026.03 | 6.88 | — | — | — | |
| KarnakSize=40B2026.03 | 6.51 | — | — | — | |
| Fanar-27BSize=27B2026.03 | 6.12 | — | — | — | |
| AceGPT-v2-70B-ChatSize=70B2026.03 | 6.01 | — | — | — | |
| Jais-2-70B-ChatSize=70B2026.03 | 5.63 | — | — | — | |
| Fanar-1-9B-InstructSize=9B2026.03 | 5.58 | — | — | — | |
| DPOBackbone=Qwen3-8B2026.03 | 5.52 | — | — | — | |
| DPOBackbone=Gemma-2-9B2026.03 | 5.05 | — | — | — | |
| DSPABackbone=Gemma-2-9B2026.03 | 5.02 | — | — | — | |
| DSPABackbone=Qwen3-8B2026.03 | 4.86 | — | — | — | |
| Allam-7B-Instruct-preview-v2Size=7B2026.03 | 4.62 | — | — | — | |
| Prompt EngBackbone=Gemma-2-9B2026.03 | 4.6 | — | — | — | |
| RepEBackbone=Qwen3-8B2026.03 | 4.48 | — | — | — | |
| DSPABackbone=Gemma-2-2B2026.03 | 4.39 | — | — | — | |
| Base ModelBackbone=Gemma-2-9B2026.03 | 4.39 | — | — | — | |
| DPOBackbone=Gemma-2-2B2026.03 | 4.36 | — | — | — | |
| RepEBackbone=Gemma-2-9B2026.03 | 4.36 | — | — | — | |
| Static-SAEBackbone=Gemma-2-9B2026.03 | 4.33 | — | — | — | |
| RepEBackbone=Gemma-2-2B2026.03 | 4.32 | — | — | — | |
| AceGPT-v2-32B-ChatSize=32B2026.03 | 4.3 | — | — | — | |
| Base ModelBackbone=Qwen3-8B2026.03 | 4.12 | — | — | — | |
| Base ModelBackbone=Gemma-2-2B2026.03 | 3.91 | — | — | — | |
| Prompt EngBackbone=Qwen3-8B2026.03 | 3.88 | — | — | — | |
| Prompt EngBackbone=Gemma-2-2B2026.03 | 3.84 | — | — | — | |
| Static-SAEBackbone=Gemma-2-2B2026.03 | 3.49 | — | — | — | |
| Static-SAEBackbone=Qwen3-8B2026.03 | 3.24 | — | — | — | |
| AdaLoRAPEFT Method=AdaLoRA, Base Model=LLaMA-2-7B, Fine-tuning Dataset=WizardLM-Evol-Instruct2024.08 | — | — | — | 4.79 | |
| AtteNTModel=LLaMA 2-7B2026.02 | — | 4.49 | 2.132 | — | |
| AtteNTModel=Mistral-7B2026.02 | — | 5.32 | 1.802 | — | |
| AtteNTModel=Gemma-7B2026.02 | — | 5.44 | 2.012 | — | |
| BA-LoRAPEFT Method=BA-LoRA, Base Model=LLaMA-2-7B, Fine-tuning Dataset=WizardLM-Evol-Instruct2024.08 | — | — | — | 5.11 | |
| CorDAPEFT Method=CorDA, Base Model=LLaMA-2-7B, Fine-tuning Dataset=WizardLM-Evol-Instruct2024.08 | — | — | — | 5.15 | |
| CorDA++PEFT Method=CorDA++, Base Model=LLaMA-2-7B, Fine-tuning Dataset=WizardLM-Evol-Instruct2024.08 | — | — | — | 5.64 | |
| DoRAPEFT Method=DoRA, Base Model=LLaMA-2-7B, Fine-tuning Dataset=WizardLM-Evol-Instruct2024.08 | — | — | — | 4.48 | |
| Full FTPEFT Method=Full Fine-tuning, Base Model=LLaMA-2-7B, Fine-tuning Dataset=WizardLM-Evol-Instruct2024.08 | — | — | — | 4.85 | |
| LoRAPEFT Method=LoRA, Base Model=LLaMA-2-7B, Fine-tuning Dataset=WizardLM-Evol-Instruct2024.08 | — | — | — | 4.6 | |
| LoRA-FAPEFT Method=LoRA-FA, Base Model=LLaMA-2-7B, Fine-tuning Dataset=WizardLM-Evol-Instruct2024.08 | — | — | — | 4.67 | |
| LoRA-GAPEFT Method=LoRA-GA, Base Model=LLaMA-2-7B, Fine-tuning Dataset=WizardLM-Evol-Instruct2024.08 | — | — | — | 5.04 | |
| LoRA+PEFT Method=LoRA+, Base Model=LLaMA-2-7B, Fine-tuning Dataset=WizardLM-Evol-Instruct2024.08 | — | — | — | 5.11 | |
| MiLoRAPEFT Method=MiLoRA, Base Model=LLaMA-2-7B, Fine-tuning Dataset=WizardLM-Evol-Instruct2024.08 | — | — | — | 4.5 | |
| PiSSAPEFT Method=PiSSA, Base Model=LLaMA-2-7B, Fine-tuning Dataset=WizardLM-Evol-Instruct2024.08 | — | — | — | 4.92 | |
| Standard Fine-tuningModel=LLaMA 2-7B2026.02 | — | 4.58 | 2.461 | — | |
| Standard Fine-tuningModel=Mistral-7B2026.02 | — | 5.03 | 2.042 | — | |
| Standard Fine-tuningModel=Gemma-7B2026.02 | — | 5.42 | 2.282 | — |