List your company
ResearchFormerly WizWand

Instruction Following on MulDimIF

84.7ScoreHeRL
32.49246.04659.673.154Mar 20, 2026Mar 22, 2026Mar 24, 2026Mar 27, 2026Mar 29, 2026Mar 31, 2026Apr 3, 2026
Updated 7mo ago

Evaluation results

MethodLinks
HeRLBackbone=Llama-3.2-3B-Instruct, Training Method=HeRL2026.0384.7
HeRLBackbone=Qwen2.5-7B-Instruct, Training Method=HeRL2026.0383.4
HeRLBackbone=Qwen3-4B-Instruct-2507, Training Method=HeRL2026.0382.5
RLVRBackbone=Qwen3-4B-Instruct-2507, Training Method=RLVR2026.0379
RLVRBackbone=Llama-3.2-3B-Instruct, Training Method=RLVR2026.0377.6
RTT-AON (Ours)Backbone=Qwen3-4B-Instruct2026.0476.75
RTT-CSR (Ours)Backbone=Qwen3-4B-Instruct2026.0476.33
RL-CSRBackbone=Qwen3-4B-Instruct2026.0474.38
RL-AONBackbone=Qwen3-4B-Instruct2026.0474.25
RLVRBackbone=Qwen2.5-7B-Instruct, Training Method=RLVR2026.0373.5
RTT-AON (Ours)Backbone=Qwen2.5-7B-Instruct2026.0471.83
RTT-CSR (Ours)Backbone=Qwen2.5-7B-Instruct2026.0469.67
RL-AONBackbone=Qwen2.5-7B-Instruct2026.0469.41
RTT-AON (Ours)Backbone=Llama3.2-3B-Instruct2026.0468.58
SFTBackbone=Qwen2.5-7B-Instruct, Training Method=SFT2026.0367.8
SFTBackbone=Llama-3.2-3B-Instruct, Training Method=SFT2026.0366.9
RL-AONBackbone=Llama3.2-3B-Instruct2026.0466.83
SFTBackbone=Qwen3-4B-Instruct-2507, Training Method=SFT2026.0366.8
RL-CSRBackbone=Qwen2.5-7B-Instruct2026.0466.67
RTT-CSR (Ours)Backbone=Llama3.2-3B-Instruct2026.0466.33
DPOBackbone=Qwen3-4B-Instruct-2507, Training Method=DPO2026.0361.5
Qwen3-4B-Instruct-2507Backbone=Qwen3-4B-Instruct-2507, Training Method=Initial2026.0357.3
RL-CSRBackbone=Llama3.2-3B-Instruct2026.0456.92
DPOBackbone=Qwen2.5-7B-Instruct, Training Method=DPO2026.0356.5
SFTBackbone=Qwen3-4B-Instruct2026.0456.5
Qwen3-4B-Instruct†Backbone=Qwen3-4B-Instruct2026.0456.08
DPOBackbone=Qwen3-4B-Instruct2026.0455.83
DPOBackbone=Llama-3.2-3B-Instruct, Training Method=DPO2026.0354.4
SFTBackbone=Qwen2.5-7B-Instruct2026.0452.83
Qwen2.5-7B-InstructBackbone=Qwen2.5-7B-Instruct, Training Method=Initial2026.0351.4
Qwen2.5-7B-Instruct†Backbone=Qwen2.5-7B-Instruct2026.0451.08
DPOBackbone=Qwen2.5-7B-Instruct2026.0451.08
SFTBackbone=Llama3.2-3B-Instruct2026.0436.5
Llama-3.2-3B-InstructBackbone=Llama-3.2-3B-Instruct, Training Method=Initial2026.0335.8
Llama3.2-3B-Instruct⋆Backbone=Llama3.2-3B-Instruct2026.0435.42
DPOBackbone=Llama3.2-3B-Instruct2026.0434.5