Qwen2.5-1.5B-Instruct
A small instruction-tuned model in the Qwen2.5 series with 1.5 billion parameters, suited to lightweight deployment and connectivity testing.
Timeline (last 90 days)
The team chose Qwen2.5-1.5B-Instruct as the language model because the hackathon schedule was tight and hardware was limited, training a model from scratch was unrealistic, and they wanted to focus their effort on the memory architecture.
Phase 1 used an RTX PRO 4000 24GB to run Qwen2.5-1.5B-Instruct deployed via vLLM, with one reply shown in Open WebUI.
In Phase 1, the author asked a question in Japanese and the model returned "Connected successfully." in English.
All three open-source models tested produced valid JSON on every test case, with a parse success rate of 1.00.
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 (5)
The team chose Qwen2.5-1.5B-Instruct as the language model because the hackathon schedule was tight and hardware was limited, training a model from scratch was unrealistic, and they wanted to focus their effort on the memory architecture.
50%UnverifiedAll three open-source models tested produced valid JSON on every test case, with a parse success rate of 1.00.
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%UnverifiedPhase 1 used an RTX PRO 4000 24GB to run Qwen2.5-1.5B-Instruct deployed via vLLM, with one reply shown in Open WebUI.
50%UnverifiedIn Phase 1, the author asked a question in Japanese and the model returned "Connected successfully." in English.
50%