TE
科技回声
首页24小时热榜最新最佳问答展示工作
GitHubTwitter
首页

科技回声

基于 Next.js 构建的科技新闻平台,提供全球科技新闻和讨论内容。

GitHubTwitter

首页

首页最新最佳问答展示工作

资源链接

HackerNews API原版 HackerNewsNext.js

© 2025 科技回声. 版权所有。

Running your models in production with TensorFlow Serving

170 点作者 hurrycane超过 9 年前

5 条评论

Smerity超过 9 年前
Model serving in production is a persistent pain point for many ML backends, and is usually done quite poorly, so this is great to see.<p>I&#x27;m expecting large leaps and bounds for TensorFlow itself. This improvement to surrounding infrastructure is a nice surprise, just as TensorBoard is one of the nicest &quot;value-adds&quot; that the original library had[4].<p>Google have ensured many high quality people have been active as evangelists[3], helping build a strong community and answerbase. While there are still gaps in what the whitepaper[1] promises and what has made it to the open source world[2], it&#x27;s coming along steadily.<p>My largest interests continue to be single machine performance (a profiler for performance analysis + speedier RNN implementations) and multi-device &#x2F; distributed execution. Single machine performance had a huge bump from v0.5 to v0.6 for CNNs, eliminating one of the pain points there, so they&#x27;re on their way.<p>I&#x27;d have expected this to lead to an integration with Google Compute Engine (TensorFlow training &#x2F; prediction as a service) except for the conspicuous lack of GPU instances on GCE. While GPUs are usually essential for training (and theoretically could be abstracted away behind a magical GCE TF layer) there are still many situations in which you&#x27;d want access to the GPU itself, particularly as performance can be unpredictable across even similar hardware and machine learning model architectures.<p>[1]: <a href="http:&#x2F;&#x2F;download.tensorflow.org&#x2F;paper&#x2F;whitepaper2015.pdf" rel="nofollow">http:&#x2F;&#x2F;download.tensorflow.org&#x2F;paper&#x2F;whitepaper2015.pdf</a><p>[2]: Extricating TensorFlow from &quot;Google internal&quot; must be a real challenge given TF distributed training interacts with various internal infra tools and there are gaps with open source equivalents.<p>[3]: Shout out to @mrry who seems to have his fingers permanently poised above the keyboard - <a href="http:&#x2F;&#x2F;stackoverflow.com&#x2F;users&#x2F;3574081&#x2F;mrry?tab=answers&amp;sort=newest" rel="nofollow">http:&#x2F;&#x2F;stackoverflow.com&#x2F;users&#x2F;3574081&#x2F;mrry?tab=answers&amp;sort...</a><p>[4]: I&#x27;ve been working on a dynamic memory network (<a href="http:&#x2F;&#x2F;arxiv.org&#x2F;abs&#x2F;1506.07285" rel="nofollow">http:&#x2F;&#x2F;arxiv.org&#x2F;abs&#x2F;1506.07285</a>) implementation recently and it&#x27;s just lovely to see a near perfect visualization of the model architecture by default - <a href="http:&#x2F;&#x2F;imgur.com&#x2F;a&#x2F;PbIMI" rel="nofollow">http:&#x2F;&#x2F;imgur.com&#x2F;a&#x2F;PbIMI</a>
评论 #11114641 未加载
评论 #11114206 未加载
评论 #11114542 未加载
dgacmu超过 9 年前
Note also that we&#x27;ve released v0.7 of Tensorflow today - more details in the release announcement: <a href="https:&#x2F;&#x2F;groups.google.com&#x2F;a&#x2F;tensorflow.org&#x2F;forum&#x2F;#!topic&#x2F;discuss&#x2F;_IdeX4XCRqg" rel="nofollow">https:&#x2F;&#x2F;groups.google.com&#x2F;a&#x2F;tensorflow.org&#x2F;forum&#x2F;#!topic&#x2F;dis...</a>
TheGuyWhoCodes超过 9 年前
This looks great and brings TensorFlow close to using it in production where the model has a life cycle.<p>I&#x27;d wish they could implement other well know ML algos like trees, give Spark ML some fight :)
评论 #11114663 未加载
swah超过 9 年前
Off-topic: I always open C++ projects from Google - they are always so tidy and clean. It just feels like a work of craftmanship, if that actually exists in software: <a href="https:&#x2F;&#x2F;github.com&#x2F;tensorflow&#x2F;serving&#x2F;tree&#x2F;master&#x2F;tensorflow_serving&#x2F;core" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;tensorflow&#x2F;serving&#x2F;tree&#x2F;master&#x2F;tensorflow...</a><p>OTOH, I have a strong prejudice against Javascript on the backend... And its not due to it being dynamic - the same doesn&#x27;t happen with Python codebases. It is completely irrational.
评论 #11116828 未加载
curiousfiddler超过 9 年前
I&#x27;m not sure if TensorFlow already provides that, but it would also be pretty awesome to access some of Google&#x27;s data sets to train the models.
评论 #11114812 未加载
评论 #11114647 未加载