harshal3099/apex-food-rd-chatml-v3-flavour
收藏资源简介:
Apex Food R&D ChatML v3数据集是v2数据集的扩展版本,新增了第12项能力:风味与味觉系统设计。该数据集涵盖了印度风味调色板设计、甜度调节、苦味掩蔽系统、酸甜平衡、香料与风味搭配、乳制品与水风味差异、天然风味系统、风味前/中/基调、粉末中的风味释放、甜叶菊/罗汉果的后味控制、可可/咖啡/麦芽/水果/香料风味架构、儿童/成人/侨民口味偏好、感官小组评分系统、享乐测试、JAR量表测试、描述性分析、保质期内的风味稳定性、风味氧化与包装相互作用、以及益生菌/蘑菇/藻类/辣木/南非醉茄/豌豆蛋白/小米等功能性成分的掩蔽。数据集包含15,000个示例,其中3,000个专门用于风味/味觉设计,格式为ChatML messages。推荐使用Qwen/Qwen3-4B作为基础模型。数据集适用于SFT响应风格、食品配方推理、风味/感官设计推理和法规引用行为,但不替代产品上市前的直接法律/法规验证。
The Apex Food R&D ChatML v3 dataset extends the v2 dataset by adding a dedicated 12th capability: Flavour & Taste System Design. This new capability covers Indian flavour palette design, sweetness modulation, bitterness masking systems, acid-sweet balance, spice-flavour pairing, dairy vs water flavour differences, natural flavour systems, flavour top/middle/base notes, flavour release in powders, aftertaste control for stevia/monk fruit, cocoa/coffee/malt/fruit/spice flavour architectures, children vs adults vs diaspora taste preference, sensory panel scoring systems, hedonic testing, JAR scale testing, descriptive analysis, flavour stability during shelf life, flavour oxidation and packaging interaction, and masking of probiotics, mushrooms, algae, moringa, ashwagandha, pea protein, millets and other functional ingredients. The dataset contains 15,000 examples, with 3,000 dedicated to flavour/taste design, formatted as ChatML messages. The recommended base model is Qwen/Qwen3-4B. The dataset is suitable for SFT response style, food formulation reasoning, flavour/sensory design reasoning and regulatory citation behaviour, but it is not a substitute for direct legal/regulatory verification before product launch.




