List your company
ResearchFormerly WizWand

Correlation Metrics on MT-Bench (LLM-as-a-Judge)

0.689Pearson's rQwen3-32B REAL (ours)
0.26260.37330.4840.5947Mar 6, 2025May 7, 2025Jul 9, 2025Sep 10, 2025Nov 11, 2025Jan 13, 2026Mar 17, 2026
Updated 7mo ago

Evaluation results

MethodLinks
Qwen3-32B REAL (ours)Training=RL, Inference=RAIL2026.030.6890.6910.552
TRACTBackbone=Llama3.1-8B, COT=true, Train=C-RAFT, Data=Self, Inf=C-RAIL2025.030.6720.639—
Baseline B.4 (RAFT)Backbone=Llama3.1-8B, COT=false, Train=RAFT, Data=GPT-4+, Inf=RAIL2025.030.6180.614—
Qwen3-8B REAL (ours)Training=RL, Inference=RAIL2026.030.6170.6080.471
Qwen3-32B RAFTTraining=SFT, Inference=RAIL2026.030.6110.5960.439
Mistral2-7B REAL (ours)Training=RL, Inference=RAIL2026.030.5930.5690.422
Qwen3-8B TRACTTraining=SFT, Inference=RAIL2026.030.5580.5860.436
TRACTBackbone=Mistral-7B-Instruct, CoT=true, Train Objective=C-RAFT, Training Data=Self, Inference Method=C-RAIL2025.030.5550.529—
Baseline B.3 (RAIL)Backbone=Llama3.1-8B, COT=false, Inf=RAIL2025.030.5470.583—
Baseline B.1 (CE)Backbone=Llama3.1-8B, COT=false, Train=CE, Data=GPT-4, Inf=Decode2025.030.5410.556—
Qwen3-8B RAFTTraining=SFT, Inference=RAIL2026.030.5410.5170.385
Prometheus-2-7BTraining=SFT, Inference=RAIL2026.030.5380.5110.389
Mistral2-7B Standard RLTraining=RL, Inference=RAIL2026.030.5290.5070.371
Mistral2-7B TRACTTraining=SFT, Inference=RAIL2026.030.5210.5010.366
Prometheus-2-7BBackbone=Prometheus-2-7B, Inference Method=Decode2025.030.5190.483—
Prometheus-2-7BTraining=SFT, Inference=Standard2026.030.5190.4830.392
TRACT (Ablation: Objective CE)Backbone=Mistral-7B-Instruct, CoT=true, Train Objective=CE, Training Data=Self, Inference Method=C-RAIL2025.030.5170.503—
CLoud (Reward model)Backbone=Llama-3-8B, Inference Method=Decode*2025.030.5110.506—
Qwen3-8B Standard RLTraining=RL, Inference=RAIL2026.030.4950.5350.397
Mistral-7B-Instruct (RAFT on GPT-4)Backbone=Mistral-7B-Instruct, CoT=false, Train Objective=RAFT†, Training Data=GPT-4†, Inference Method=RAIL2025.030.4830.469—
Mistral-7B-Instruct (CE on GPT-4 CoT)Backbone=Mistral-7B-Instruct, CoT=true, Train Objective=CE, Training Data=GPT-4, Inference Method=Decode2025.030.480.482—
Prometheus-1-13BTraining=SFT, Inference=Standard2026.030.4670.4550.345
Baseline B.2 (CE)Backbone=Llama3.1-8B, COT=true, Train=CE, Data=GPT-4, Inf=Decode2025.030.4660.494—
Mistral-7B-Instruct (CE on Self-gen CoT)Backbone=Mistral-7B-Instruct, CoT=true, Train Objective=CE, Training Data=Self, Inference Method=Decode2025.030.4350.426—
TRACT (Ablation: Stage 1 Init)Backbone=Mistral-7B-Instruct, Note=Stage 2 initialized from ps, CoT=true2025.030.4320.421—
Qwen3-32B BaseTraining=None, Inference=RAIL2026.030.4250.4680.353
GPT-3.5-Turbo-0613Training=None, Inference=Standard2026.030.4220.3710.299
TRACT (Ablation: Data GPT-4)Backbone=Mistral-7B-Instruct, CoT=true, Train Objective=C-RAFT, Training Data=GPT-4, Inference Method=C-RAIL2025.030.3990.418—
Mistral2-7B RAFTTraining=SFT, Inference=RAIL2026.030.3990.4180.307
Prometheus-1-7BTraining=SFT, Inference=Standard2026.030.3670.3710.285
Qwen3-8B BaseTraining=None, Inference=RAIL2026.030.3590.3550.27
Qwen3-8B BaseTraining=None, Inference=Standard2026.030.3550.3270.265
Mistral2-7B Base (w/ warmup)Training=None, Inference=RAIL2026.030.320.2990.214
Mistral-7B-Instruct (Zero-shot RAIL)Backbone=Mistral-7B-Instruct, CoT=false, Inference Method=RAIL2025.030.3090.216—
Mistral2-7B Base (w/ warmup)Training=None, Inference=Standard2026.030.3090.3180.253
Mistral-7B-Instruct (CE on GPT-4 scores)Backbone=Mistral-7B-Instruct, CoT=false, Train Objective=CE, Training Data=GPT-4†, Inference Method=Decode2025.030.2790.268—