其他
几行代码,GPT-3变ChatGPT!吴恩达高徒、华人CEO震撼发布Lamini引擎
新智元报道
新智元报道
【新智元导读】这个全新发布的Lamini引擎,大大拉低了模型训练的门槛,开发者人手一个ChatGPT的梦想成真了。
MLOps的未来是「LMOps」。哪里有标准化,哪里就有机会。
训LLM就像prompt-tuning一样简单
任意LLM,秒变ChatGPT!
from llama import
LLMllm = LLM(name="marketing")
from llama import Type, Context
class AdAspects(Type):
tone: str = Context("tone of the marketing copy")
product_features: list = Context("product features to promote")
audience: str = Context("target audience for the message")
subject: str = Context("subject or topic of the message")
goal: str = Context("goal of this marketing campaign and message")
class AdCopy(Type):
title: str = Context("google ad title tag")
description: str = Context("google ad description")
keywords: list = Context("keywords for the search engine")
语气:大胆,但不傲慢 特色:亚洲酱料和香料、家常调料和套餐包,可以轻松在家烹饪。
aspects = AdAspects(
tone="bold and bright, but not arrogant",
product_features=[
'asian sauces and aromatics',
'home-cooked seasonings and meal packs that can be easily cooked at home'
],
audience="suburban families",
subject="delicious asian meals without going to a restaurant",
goal="get suburban moms and dads to try buy their first omsom pack or free tasting kit"
)
ad_copy = llm(input=aspects, output_type=AdCopy)
print(f"Ad copy: {ad_copy}")
尝试 Omsom 的美味亚洲酱料、香料、家常调料和套餐包。轻松为家人在家做出美味佳肴。
> title='Delicious Asian Meals Without Going to a Restaurant | Omsom'
description="Try Omsom's delicious Asian sauces, aromatics, and home-cooked seasonings and meal packs. Easily cook delicious meals at home for your family."
keywords=[
'Asian sauces',
'Aromatics',
'Home-cooked seasonings',
'Meal packs',
'Delicious meals',
'Suburban families',
'Omsom'
]
如何创建自己的「ChatGPT」
尝试prompt-tuning ChatGPT或其他模型
构建一个包含输入-输出对的大型数据集
在大型数据集上微调基础模型
在微调模型上进行RLHF
部署到云端
专为LLM打造的数据生成器
用于优化prompt微调和类型化输出(typed outputs )的 Lamini库。 用于微调和RLHF的高级Lamini库,只需几行代码。 史上首个托管数据生成器,用于创建数据,来训练遵循指令的LLM。注意,已获得商业使用许可! 开源的指令跟随(instruction-following)LLM,使用上述工具,只需几行代码即可完成。
数据生成器工作原理
对生成数据进行微调
团队介绍