List your company

Mathematical Reasoning on GSM8K (test) (Accuracy, #Token (k))

89.2AccuracyCARD
18.27236.68655.173.514Dec 24, 2024Mar 2, 2025May 9, 2025Jul 16, 2025Sep 22, 2025Nov 29, 2025Feb 5, 2026

Evaluation results

MethodLinks
CARDBackbone=Qwen3-1.7B2025.1289.2——921
Qwen3-4B-Inst-2507Chat Template=Off, Decoding=Static, k=12026.0289.11——
Qwen3-4B-Inst-2507Chat Template=Off, Decoding=ConfAdapt, tau=0.9952026.0288.22.1——
RDoLTBackbone=Qwen3-1.7B, Strategy=3-tier2025.1287.9——1,987
Qwen3-4B-Inst-2507Chat Template=Off, Decoding=ConfAdapt, tau=0.992026.0287.62.2——
Qwen3-4B-Inst-2507Chat Template=Off, Decoding=ConfAdapt, tau=0.982026.0287.52.4——
Qwen3-4B-Inst-2507Chat Template=Off, Decoding=ConfAdapt, tau=0.972026.0287.52.5——
Zero-ShotBackbone=Qwen3-1.7B2025.1287.5——1,798
Qwen3-4B-Inst-2507Chat Template=Off, Decoding=ConfAdapt, tau=0.962026.02872.6——
Qwen3-4B-Inst-2507Chat Template=Off, Decoding=ConfAdapt, tau=0.952026.0286.92.7——
ToTBackbone=Qwen3-1.7B, Search strategy=beam-32025.1286.8——1,542
CARDBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.1285.8——1,043
MCTS w/ Var Reducesearch_strategy=MCTS, variant=Var Reduce2025.0285.48.84——
MCTS w/ FETCHsearch_strategy=MCTS, variant=FETCH2025.0285.32.6——
BFS w/ FETCHsearch_strategy=BFS, variant=FETCH2025.0285.20.87——
CoTBackbone=Qwen3-1.7B2025.1285.1——816
BFS w/ Var Reducesearch_strategy=BFS, variant=Var Reduce2025.0284.72.25——
Beam Search w/ Var Reducesearch_strategy=Beam Search, variant=Var Reduce2025.0284.61.58——
MCTS w/ State Mergesearch_strategy=MCTS, variant=State Merge2025.0284.53.03——
Beam Search w/ FETCHsearch_strategy=Beam Search, variant=FETCH2025.0284.21.19——
RDoLTBackbone=DeepSeek-R1-Distill-Qwen-1.5B, Strategy=3-tier2025.1284.1——2,289
Tree Search w/ Var Reducesearch_strategy=Tree Search, variant=Var Reduce2025.0283.80.53——
MCTSsearch_strategy=MCTS2025.0283.810.7——
Tree Search w/ FETCHsearch_strategy=Tree Search, variant=FETCH2025.0283.70.46——
Zero-ShotBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.1283.65——2,043
BFS w/ State Mergesearch_strategy=BFS, variant=State Merge2025.0283.60.88——
Qwen3-4B-Inst-2507Chat Template=Off, Decoding=ConfAdapt, tau=0.92026.0283.63.1——
Beam Search w/ State Mergesearch_strategy=Beam Search, variant=State Merge2025.0283.21.17——
ToTBackbone=DeepSeek-R1-Distill-Qwen-1.5B, Search strategy=beam-32025.1282.9——3,187
Beam Searchsearch_strategy=Beam Search2025.0282.81.55——
Tree Search w/ State Mergesearch_strategy=Tree Search, variant=State Merge2025.0282.50.48——
BFSsearch_strategy=BFS2025.0282.42.47——
Qwen3-4B-Inst-2507Chat Template=Off, Decoding=Static, k=22026.0282.32——
Qwen3-4B-Inst-2507Chat Template=Off, Decoding=ConfAdapt, tau=0.872026.0282.33.2——
Best-of-Nsearch_strategy=Best-of-N2025.0281.90.83——
Tree Searchsearch_strategy=Tree Search2025.0281.80.55——
Weighted Votingsearch_strategy=Weighted Voting2025.0281.70.83——
CARDBackbone=Qwen3-0.6B2025.1281.4——897
Qwen3-4B-Inst-2507Chat Template=Off, Decoding=ConfAdapt, tau=0.852026.0281.33.3——
CoTBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.1281.2——2,066
RDoLTBackbone=Qwen3-0.6B, Strategy=3-tier2025.1279.8——2,156
Zero-ShotBackbone=Qwen3-0.6B2025.1279.2——1,687
TALE-PT-SFTLLM=Llama-3.1-8B-Instruct2024.1278.57—139.63—
ToTBackbone=Qwen3-0.6B, Search strategy=beam-32025.1278.4——2,934
Qwen3-4B-Inst-2507Chat Template=Off, Decoding=ConfAdapt, tau=0.82026.0277.63.6——
Vanilla CoTLLM=Llama-3.1-8B-Instruct2024.1277.56—241.51—
CoTBackbone=Qwen3-0.6B2025.1276.8——1,847
Qwen3-4B-Inst-2507Chat Template=Off, Decoding=Baseline, Step=0, k=12026.0275.41——
Self-Consistencysearch_strategy=Self-Consistency2025.0275.30.83——
TALE-PT-DPOLLM=Llama-3.1-8B-Instruct2024.1274.11—149.93—
Qwen3-4B-Inst-2507Chat Template=Off, Decoding=ConfAdapt, tau=0.752026.0273.23.8——
L3.1-8B-MagpieChat Template=On, Decoding=Baseline, Step=0, k=12026.0269.51——
Qwen3-4B-Inst-2507Chat Template=Off, Decoding=ConfAdapt, tau=0.72026.0268.24.1——
L3.1-8B-MagpieChat Template=On, Decoding=Static, k=12026.02661——
L3.1-8B-MagpieChat Template=On, Decoding=ConfAdapt, tau=0.992026.02662.1——
Greedy Decodingsearch_strategy=Greedy Decoding2025.0265.70.08——
L3.1-8B-MagpieChat Template=On, Decoding=ConfAdapt, tau=0.9952026.0265.71.9——
L3.1-8B-MagpieChat Template=On, Decoding=ConfAdapt, tau=0.982026.0265.52.3——
L3.1-8B-MagpieChat Template=On, Decoding=ConfAdapt, tau=0.972026.0265.32.5——
L3.1-8B-MagpieChat Template=On, Decoding=ConfAdapt, tau=0.962026.0264.92.6——
L3.1-8B-MagpieChat Template=On, Decoding=ConfAdapt, tau=0.952026.0264.72.8——
L3.1-8B-MagpieChat Template=On, Decoding=ConfAdapt, tau=0.92026.0264.13.3——
L3.1-8B-MagpieChat Template=On, Decoding=ConfAdapt, tau=0.872026.0262.33.6——
Qwen3-4B-Inst-2507Chat Template=Off, Decoding=ConfAdapt, tau=0.652026.02624.4——
Qwen3-4B-Inst-2507Chat Template=Off, Decoding=Static, k=32026.0261.63——
L3.1-8B-MagpieChat Template=On, Decoding=ConfAdapt, tau=0.852026.0261.23.7——
L3.1-8B-MagpieChat Template=On, Decoding=Static, k=22026.0260.72——
L3.1-8B-MagpieChat Template=On, Decoding=ConfAdapt, tau=0.82026.0259.24.2——
Qwen3-4B-Inst-2507Chat Template=Off, Decoding=ConfAdapt, tau=0.62026.0258.84.6——
L3.1-8B-MagpieChat Template=On, Decoding=ConfAdapt, tau=0.752026.0256.94.6——
L3.1-8B-MagpieChat Template=On, Decoding=ConfAdapt, tau=0.72026.02555.1——
L3.1-8B-MagpieChat Template=On, Decoding=Static, k=32026.02533——
L3.1-8B-MagpieChat Template=On, Decoding=ConfAdapt, tau=0.652026.0251.55.6——
L3.1-8B-MagpieChat Template=On, Decoding=Static, k=42026.0249.84——
L3.1-8B-MagpieChat Template=On, Decoding=ConfAdapt, tau=0.62026.0246.66.1——
Qwen3-4B-Inst-2507Chat Template=Off, Decoding=Static, k=42026.0244.24——
L3.1-8B-MagpieChat Template=On, Decoding=Static, k=52026.0241.55——
Qwen3-4B-Inst-2507Chat Template=Off, Decoding=Static, k=52026.0228.95——
Directly AnsweringLLM=Llama-3.1-8B-Instruct2024.1221—38.54—