List your company

Math on GSM8K 1,000-example (test)

5.8317PPLQwen3-VL-2B-Instruct
5.3901568.37057811.35114.331422Jan 13, 2026

Evaluation results

MethodLinks
Qwen3-VL-2B-InstructAblation State=Base, Evaluation Mode=Teacher-forced, Sequence Length=Full-sequence, Seed=422026.015.8317—
Qwen3-VL-2B-InstructAblation State=SRA, Evaluation Mode=Teacher-forced, Sequence Length=Full-sequence, Seed=422026.015.87560.0439
Qwen3-VL-8B-InstructAblation State=SRA, Evaluation Mode=Teacher-forced, Sequence Length=Full-sequence, Seed=422026.016.5793-0.1754
Qwen3-VL-8B-InstructAblation State=Base, Evaluation Mode=Teacher-forced, Sequence Length=Full-sequence, Seed=422026.016.7547—
Qwen3-VL-4B-InstructAblation State=Base, Evaluation Mode=Teacher-forced, Sequence Length=Full-sequence, Seed=422026.017.0319—
Qwen3-VL-4B-InstructAblation State=SRA, Evaluation Mode=Teacher-forced, Sequence Length=Full-sequence, Seed=422026.017.06910.0372
Ministral-14B-Instruct-2512Ablation State=SRA, Evaluation Mode=Teacher-forced, Sequence Length=Full-sequence, Seed=422026.019.1063-0.2704
Ministral-14B-Instruct-2512Ablation State=Base, Evaluation Mode=Teacher-forced, Sequence Length=Full-sequence, Seed=422026.019.3767—
Ministral-3B-Instruct-2512Ablation State=SRA, Evaluation Mode=Teacher-forced, Sequence Length=Full-sequence, Seed=422026.0116.1005-0.7698
Ministral-3B-Instruct-2512Ablation State=Base, Evaluation Mode=Teacher-forced, Sequence Length=Full-sequence, Seed=422026.0116.8703—