Keepace.ai
A fine-tuned large model product focused on Chinese-language fitness and health, post-trained from Qwen3.5-27B. It covers three task categories: training plan generation, fitness and health knowledge question answering, and personal exercise data analysis.
Timeline (last 90 days)
Keepace.ai performs post-training based on Qwen3.5-27B, focusing on Chinese exercise guidance.
Keepace.ai covers three core task types: generating training plans, answering exercise and health knowledge questions, and analyzing personal exercise data.
In an evaluation designed by internal experts, Keepace.ai outperforms the base model Qwen3.5-27B and larger general-purpose models on multiple metrics.
Keepace.ai's evaluation system was designed with the participation of domain experts and uses rubric-based internal evaluation, with particular emphasis on exercise safety and boundary compliance.
Compared with the base model and larger general-purpose models, Keepace.ai shows the most pronounced improvements in four dimensions: exercise safety, boundary compliance, structural output completeness, and scientific accuracy.
The Keepace.ai technical report does not disclose details of the specific post-training methods (such as SFT, RLHF, DPO, etc.).
The Keepace.ai technical report does not disclose the specific scale and source composition of the training data, nor the quantitative scores of the evaluation.
Keepace.ai's current optimization is limited to Chinese scenarios, and multilingual capability has not yet been addressed.
Keepace.ai's evaluation consists of internal expert scoring and lacks third-party validation on public benchmarks.
Keepace.ai's analysis holds that its "safety-first" evaluation approach is more worth learning from by AI practitioners in vertical domains than the model itself.
All Facts (10)
Keepace.ai performs post-training based on Qwen3.5-27B, focusing on Chinese exercise guidance.
50%UnverifiedKeepace.ai covers three core task types: generating training plans, answering exercise and health knowledge questions, and analyzing personal exercise data.
50%UnverifiedIn an evaluation designed by internal experts, Keepace.ai outperforms the base model Qwen3.5-27B and larger general-purpose models on multiple metrics.
50%UnverifiedKeepace.ai's evaluation system was designed with the participation of domain experts and uses rubric-based internal evaluation, with particular emphasis on exercise safety and boundary compliance.
50%UnverifiedCompared with the base model and larger general-purpose models, Keepace.ai shows the most pronounced improvements in four dimensions: exercise safety, boundary compliance, structural output completeness, and scientific accuracy.
50%UnverifiedThe Keepace.ai technical report does not disclose details of the specific post-training methods (such as SFT, RLHF, DPO, etc.).
50%UnverifiedThe Keepace.ai technical report does not disclose the specific scale and source composition of the training data, nor the quantitative scores of the evaluation.
50%UnverifiedKeepace.ai's current optimization is limited to Chinese scenarios, and multilingual capability has not yet been addressed.
50%UnverifiedKeepace.ai's evaluation consists of internal expert scoring and lacks third-party validation on public benchmarks.
50%UnverifiedKeepace.ai's analysis holds that its "safety-first" evaluation approach is more worth learning from by AI practitioners in vertical domains than the model itself.
50%