Видео

Stroman Walks

Kunze Junctions

Declan Mews

Lebsack Light

Tyrell Mountain

Ella Burgs

470 White Branch

Thurman Terrace

Inspiring talk how to bring RL into industrial practice, thanks for sharing!

Lenora Isle

Ettie Road

Fae Harbors

Excellent talk

Magnus Ridges

Thank you very much for this !

this talk can't be more good. Thanks Hao!!!

This was a really insightful presentation. It made me think about how different tools approach information retrieval. I've been using Myko Assistant recently, and its deep internet search feature really helps me find accurate answers quickly, especially compared to some others like Perplexity.

is the PDF all text? Does it have images, tables, graphs? And if so, how well does it convert?

Impressive

It was great. Thank you!

The first code slide contains an error. Profing your code example when aiming at software engineers is important :)

Great vid

no thanks

Thanks for explaining simple, I migrated to RLlib and it's really efficient.

Great presentation. Just one question: What is relevance_score in this case? Is it an aggregation of grounding metrics for all reference examples?

very helpful and distinguish presentation!

It should be noted, that since this talk, Anyscale deprecated Ray LLM and now recommend vLLM

Hi! What are the chances we can get the source code for this demo!! Trying to learn about KubeRay myself, have a k8s cluster (with gpu and taints setup, but cant get kuberay to play nicely with them. my job just gets stuck w/ pending)

great content! thanks a lot!

great video, that's a lot of useful features !

How to protect a company's information with technology ?

Thanks but what about scanned pdfs any way to handle the exceptions

By YouSum Live 00:00:11 Future of search innovation. 00:01:17 Linear story vs. reality of innovation. 00:02:37 Building solutions for personal challenges. 00:05:28 Importance of personal product usage. 00:06:02 Learning through personal experiences. 00:09:19 Iterative improvement and momentum. 00:10:34 Enhancing search with live knowledge integration. 00:13:20 Introduction of Copilot for interactive browsing. 00:18:06 Transition to a comprehensive research platform. 00:19:49 Importance of orchestration in complex systems. 00:20:41 Challenges in plugin and API integration reliability. 00:21:21 Evaluation metrics for generative search engines. 00:23:01 Continuous iteration and improvement for product success. 00:23:22 Fusion of Wikipedia and chat for deep topic exploration. 00:23:55 Development of faster and efficient llama models. 00:24:42 Customization of models for improved performance. 00:30:25 Balancing between open-source and proprietary models. 00:31:19 Control over pricing and business perspective in model development. By YouSum Live

If my models are unrelated and have no functional requirements to run together in a single application, can I still use Model composition in Ray serve to deploy multiple model in a single application providing a unified API endpoint (with different route for each model) for better resource utilisation and easier deployment? Is it a good practice? What about the security aspects and user authentications?

I love the way how he shared the story

How do we calculate memory used by kv cache in paged attention.Example for input 500 and output 1000

you are reading what is on the screen, do you think that you are "useful" ? or "informative "??! not at all

Great talk and amazing work guys!

Neat.

Interesting sharings. Curious about the underlying implementation for KV blocks sharing part you have a copy-on-write mechanism, but how does it avoid dirty-read condition, where both request reads that ref count is 2 and both request copies the block simultaneously.

great work

Great explanation!

Good content, thanks for your research

When I use ray.init() it starts both the dashboard and a job at the same time. Is there a way to start the dashboard, then a separate command to start individual jobs?

Considering the 300g ray vs spark comparison (~15m30s-18m30s) - the spark side seems to save all the prediction outputs (`...write...save()`), but I don't see that on the ray side (`for _ in ds.iter_batches(..: pass`). Does ray's 'iter_batches()` automatically dump outputs somewhere? (e.g. when specifying `batch_format='pyarrow'` does it get automatically cached or sth in the ray object store, or sth similar?) If not - I'd argue it's not an entirely fair apples-to-apples comparison?

I really like your approach, some libraries like keras are difficult to use

Excellent, very educating

First Impressions! So human. :)

awesome talk, with useful insights!