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。

Show HN: OpenLLMetry – OpenTelemetry-based observability for LLMs

154 点作者 nirga超过 1 年前
Hey HN, Nir, Gal and Tomer here. We’re open-sourcing a set of extensions we’ve built on top of OpenTelemetry that provide visibility into LLM applications - whether it be prompts, vector DBs and more. Here’s the repo: <a href="https:&#x2F;&#x2F;github.com&#x2F;traceloop&#x2F;openllmetry">https:&#x2F;&#x2F;github.com&#x2F;traceloop&#x2F;openllmetry</a>.<p>There’s already a decent number of tools for LLM observability, some open-source and some not. But what we found was missing for all of them is that they were closed-protocol by design, vendor-locking you to use their observability platform or their proprietary framework for running your LLMs.<p>It’s still early in the gen-AI space so we think it’s the right time to define an open protocol for observability. So we built OpenLLMetry. It extends OpenTelemetry and provides instrumentations for LLM-specific libraries which automatically monitor and trace prompts, token usage, embeddings, etc.<p>Two key benefits with OpenTelemetry are (1) you can trace your entire system execution, not just the LLM (so you can see how requests to DBs, or other calls affect the overall result); and (2) you can connect to any monitoring platform—no need to adopt new tools. Install the SDK and plug it into Datadog, Sentry, or both. Or switch between them easily.<p>We’ve already built instrumentations for LLMs like OpenAI, Anthropic and Cohere, vector DBs like Pinecone and LLM Frameworks like LangChain and Haystack. And we’ve built an SDK that makes it easy to use all of these instrumentations in case you’re not too familiar with OpenTelemetry.<p>Everything is written in Python (with Typescript around the corner) and licensed with Apache-2.0.<p>We’re using this SDK for our own platform (Traceloop), but our hope is that OpenLLMetry can evolve and thrive independently, giving everyone (including our users) the power of choice. We’ll be working with the OpenTelemetry community to get this to become a first-class citizen of OpenTelemetry.<p>Would love to hear your thoughts and opinions!<p>Check it out -<p>Docs: <a href="https:&#x2F;&#x2F;www.traceloop.com&#x2F;docs&#x2F;python-sdk&#x2F;introduction">https:&#x2F;&#x2F;www.traceloop.com&#x2F;docs&#x2F;python-sdk&#x2F;introduction</a><p>Github: <a href="https:&#x2F;&#x2F;github.com&#x2F;traceloop&#x2F;openllmetry">https:&#x2F;&#x2F;github.com&#x2F;traceloop&#x2F;openllmetry</a>

15 条评论

Areibman超过 1 年前
LLM observability strikes me as an extremely, extremely crowded space. And YC has funded an enormous number of them.<p>What do you think is the key differentiator between you and everyone else? Is vendor lock-in really that huge of an issue?<p>[0] <a href="https:&#x2F;&#x2F;hegel-ai.com">https:&#x2F;&#x2F;hegel-ai.com</a>, <a href="https:&#x2F;&#x2F;www.vellum.ai&#x2F;">https:&#x2F;&#x2F;www.vellum.ai&#x2F;</a>, <a href="https:&#x2F;&#x2F;www.parea.ai">https:&#x2F;&#x2F;www.parea.ai</a>, <a href="http:&#x2F;&#x2F;baserun.ai">http:&#x2F;&#x2F;baserun.ai</a>, <a href="https:&#x2F;&#x2F;www.trychatter.ai">https:&#x2F;&#x2F;www.trychatter.ai</a>, <a href="https:&#x2F;&#x2F;talc.ai">https:&#x2F;&#x2F;talc.ai</a>, <a href="https:&#x2F;&#x2F;github.com&#x2F;BerriAI&#x2F;bettertest">https:&#x2F;&#x2F;github.com&#x2F;BerriAI&#x2F;bettertest</a>, <a href="https:&#x2F;&#x2F;langfuse.com">https:&#x2F;&#x2F;langfuse.com</a>
评论 #37850576 未加载
评论 #37852321 未加载
caniszczyk超过 1 年前
Any thoughts of contributing this upstream directly or to CNCF?<p>We would be interested in hosting and supporting this type of work.<p>You can reach out to me via cra@linuxfoundation.org if you want to chat
评论 #37850602 未加载
ramenmeal超过 1 年前
What is the difference from using OpenLLMetry versus using OTel directly? Is the issue that there aren&#x27;t conventions for the needed attributes?
评论 #37847960 未加载
jeffchao超过 1 年前
Cool! It looks like you effectively do auto instrumentation. Have you found there to be interesting nuances between LLM providers? Tracing is great and trace aggrgegates (with context!) cross-vendor would be even more awesome.
评论 #37850274 未加载
serverlessmom超过 1 年前
Pretty neat! I assume it&#x27;s just measuring traces right now? Any plans to add some top level metrics like build times, prompt length, etc?
评论 #37849501 未加载
hossko超过 1 年前
Hello,<p>Is it possible to use Traceloop&#x27;s LLM instrumentations with already existing opentelemetry implementation ?
评论 #37854025 未加载
hossam-shehab超过 1 年前
Hey,<p>Is it possible to used Traceloop LLM instrumentations only with already existing opentelemetry implementation
peter_d_sherman超过 1 年前
Great idea!<p>Observability (AKA, debug&#x2F;proxy&#x2F;statistics&#x2F;logging&#x2F;visualization layer) -- for LLM&#x27;s (AKA Chat AI&#x27;s)...<p>Hmmm, you know, I would love something for ChatGPT (and other AI chatbots) -- where you could open a second tab or window -- and see (and potentially interact with) debug info and statistics from prompts given to that AI in its main input window, in realtime...<p>Sort of like what Unix&#x27;s STDERR is for programs running on Unix -- but an &quot;AI STDERR&quot; AKA debug channel, for AI&#x27;s...<p>I&#x27;m guessing (but not knowing) that in the future, there will be standards defined for debug interfaces to AI&#x27;s, standards defined for the data formats and protocols traversing those interfaces, and standards defined for such things as error, warning, hint, and informational messages...<p>Oh sure, a given AI company could pick a series of their own interfaces, data protocols and how to interpret that data.<p>But if so, that &quot;AI debug interface&quot; -- wouldn&#x27;t be universal.<p>Of course, on the flip side, if a universal &quot;AI debug interface&quot; were ever established, perhaps such a thing would eventually suffer from the complexities, over-engineering and bloatedness that plague many &quot;designed-by-committee&quot; standards in today&#x27;s world.<p>So, it will be interesting to see what the future holds...<p>To take an Elon Musk quote and twist it around (basically abuse it! &lt;g&gt;):<p>&quot;Proper engineering of future designed-by-committee standards with respect to AI interfaces and protocols is NOT guaranteed -- but excitement is!&quot;<p>:-) &lt;g&gt; :-)<p>Anyway, with respect to the main subject&#x2F;article&#x2F;authors, it&#x27;s a very interesting and future-thinking idea what you&#x27;re doing, you&#x27;re breaking new ground, and I wish you all of the future success with your company, business, product and product ideas!
评论 #37850530 未加载
archibaldJ超过 1 年前
any smooth way to get this work with javascript? would love to use this in a project but my inferences are all in js
评论 #37852140 未加载
brianhorakh超过 1 年前
Any plans for pgvector? Graphana tempo
评论 #37847614 未加载
VadimPR超过 1 年前
Will vLLM be supported as well?
评论 #37854180 未加载
nadavwiz超过 1 年前
Love it!
robertlagrant超过 1 年前
Would&#x27;ve preferred LLMetry, My Dear Watson.
评论 #37847848 未加载
评论 #37848587 未加载
LoganDark超过 1 年前
Worst pun ever, starred.
评论 #37847794 未加载
hartator超过 1 年前
&gt; observability<p>I really don&#x27;t like that word for some reason. It&#x27;s abstracting away something simple. Logs? Graphs? Debug data? Telemetry data? There is way better words for &quot;this&quot;.
评论 #37849202 未加载