首页
学习
活动
专区
工具
TVP
发布
精选内容/技术社群/优惠产品,尽在小程序
立即前往
您找到你想要的搜索结果了吗?
是的
没有找到

中文预训练模型泛化能力挑战赛Baseline

自从2017年具有划时代意义的Transformer模型问世以来,短短两年多的时间内,如雨后春笋般的出现了大量的预训练模型,比如:Bert,Albert,ELECTRA,RoBERta,T5,GPT3等等。然而之前的基准评测体系存在两个缺陷:评测强度不够,模型不通用。评测强度不够指的是选手只提交结果,不提交inference的代码。模型不通用指的是预训练模型不能保证在相同超参数情况下在所有任务上都获得比较好的性能。以上两点极大限制了预训练技术的应用和发展。如果我们能通过算法实现泛化能力强的中文预训练模型,这将提高下游业务的准确性,从而提升企业的核心竞争力,并为企业创造更多的价值。

02

开源基金会必须合作以防止下一次Log4Shell混乱

作为一个将整个职业生涯都花在开源软件(OSS)上的人,Log4Shell 的混乱(整个行业范围内的四级警报,以解决 Apache Log4j 包中的一个严重漏洞)是一个谦卑的提醒,我们还有很长的路要走。OSS 现在是现代社会运作的中心,就像公路桥、银行支付平台和手机网络一样重要,是时候 OSS 基金会开始像现代社会运作的中心一样行动了。 像 Apache 软件基金会、Linux 基金会、Python 基金会等组织,为他们的 OSS 开发人员社区提供法律、基础设施、营销和其他服务。在许多情况下,这些组织的安全工作资源不足,并且由于害怕吓跑新的贡献者,在设置标准和需求以减少重大漏洞的机会方面受到限制。太多的组织没有申请筹集到的资金,也没有设置过程标准来改进他们的安全实践,并且不明智地偏向于代码的数量而不是质量。 “像现代社会运作的中心一样行动”是什么样子的?以下是一些开源软件基金会可以做的事情来降低安全风险:

03

Import Kafka data into OSS using E-MapReduce service

Overview Kafka is a frequently-used message queue in open-source communities. Although Kafka (Confluent) officially provides plug-ins to import data directly from Kafka to HDFS's connector, Alibaba Cloud provides no official support for the file storage system OSS. This article will give a simple example to implement data writes from Kafka to Alibaba Cloud OSS. Because Alibaba Cloud E-MapReduce service integrates a large number of open-source components and docking tools for Alibaba Cloud, in this article, the example is directly run in the E-MapReduce cluster. This example uses the open-source Flume tool as a transit to connect Kafka and OSS. Flume open-source components may also appear on the E-MapReduce platform in the future. Scenario example Next we will name a simple example. If you already have an online Kafka cluster, you can directly jump to Step 4. 1. In the Kafka Home directory, start the Kafka service process. Configure the Zookeeper address in the configuration file to the service address emr-header-1:2181 bin/kafka-server-start.sh config/server.properties 2. Create a Kafka topic with a name of test bin/kafka-topics.sh --create --zookeeper emr-header-1:2181 \ --replication-factor 1 --partitions 1 --topic test 3. Write data to Kafka test topic and the data content is the performance monitoring data of the local machine vmstat 1 | bin/kafka-console-producer.sh --broker-list localhost:9092 --topic test 4. Configure and start the Flume service in the Flume Home directory Create a new configuration file: conf/kafka-example.conf. In specific, specify the source as the corresponding topic for Kafka, and use sink as the HDFS Sinker. Specify the path as the OSS path. Because the E-MapReduce service implements an efficient OSS FileSystem (compatible with Hadoop FileSystem) for us, the OSS path can be specified directly, and the HDFS Sinker data will be automatically written to OSS. # Name the components on this agent a1.sources = source1 a1.sinks = oss1 a1.channels = c1 # Describe/configure

03
领券