Qwen2.5-3B-Instruct
A 3-billion-parameter instruction-tuned large language model released by Alibaba's Qwen team, supporting multiple languages and long context and suited to dialogue and reasoning tasks.
Timeline (last 90 days)
The example teaches Qwen2.5-3B-Instruct to retrieve from a BM25 corpus while answering HotpotQA questions.
With a configuration of 14 rollout GPUs plus 2 trainer GPUs, after 100 training steps, exact match on the held-out set rose from 14.5% to 47%.
All three open-source models tested produced valid JSON on every test case, with a parse success rate of 1.00.
Qwen2.5-3B-Instruct achieved an overall score of 0.90 on EvidenceFlip, with an action accuracy of 0.90, conflict detection of 1.00, and pressure resistance of 1.00.
Among the three tested models, only Qwen2.5-3B can reliably refuse to answer when the evidence does not support an answer.
Parameter count is not the decisive factor in epistemic calibration ability, but within the same family there is a huge leap from 3B compared to 1.5B (0.90 vs. 0.15), indicating that scale still provides significant gains.
All Facts (6)
The example teaches Qwen2.5-3B-Instruct to retrieve from a BM25 corpus while answering HotpotQA questions.
50%UnverifiedWith a configuration of 14 rollout GPUs plus 2 trainer GPUs, after 100 training steps, exact match on the held-out set rose from 14.5% to 47%.
50%UnverifiedAll three open-source models tested produced valid JSON on every test case, with a parse success rate of 1.00.
50%UnverifiedQwen2.5-3B-Instruct achieved an overall score of 0.90 on EvidenceFlip, with an action accuracy of 0.90, conflict detection of 1.00, and pressure resistance of 1.00.
50%UnverifiedAmong the three tested models, only Qwen2.5-3B can reliably refuse to answer when the evidence does not support an answer.
50%UnverifiedParameter count is not the decisive factor in epistemic calibration ability, but within the same family there is a huge leap from 3B compared to 1.5B (0.90 vs. 0.15), indicating that scale still provides significant gains.
50%