cheptsov

9 ms·

31.
by cheptsov 2y ago
Oh, excited to see dstack featured here. Founder and core contributor to dstack here. Yes, we aim to simplify container orchestration for AI and build an alternative to both K8S and Slurm - for both multi-cloud and on-prem. Would love to he
32.
by cheptsov 2y ago
I can completely relate to anyone abandoning K8s. I'm working with dstack, an open-source alternative to K8s for AI infra [1]. We talk to many people who are frustrated with K8s, especially for GPU and AI workloads. [1] https:/&#
33.
by cheptsov 2y ago
Yes, to me, the report is interesting not because of the insights it offers, but because it raises more questions: 1. What about new AI chip vendors? 2. How will the price of compute change? 3. How will the demand for compute change? 4. How
34.
Notable AI Models(epochai.org)
25 points by cheptsov 2y ago|5 comments
35.
by cheptsov 2y ago
Unfortunately, neither VLLM nor TGI support FP8 on AMD yet. But once they do, we will look into it.
36.
by cheptsov 2y ago
TL;DR - We explore how the inference performance of Llama 3.1 405B varies on 8x AMD MI300X GPUs across vLLM and TGI backends in different use cases. - TGI is highly efficient at handling medium to high workloads. In our tests on 8x AMD MI30
37.
Developing Rapidly with Generative AI at Discord(discord.com)
1 points by cheptsov 2y ago|0 comments
38.
Intel halts plans for flagship factories in Germany, Poland amid financial woes(euractiv.com)
2 points by cheptsov 2y ago|0 comments
39.
Telegram CEO's brother also wanted by French authorities(politico.eu)
6 points by cheptsov 2y ago|1 comments
40.
by cheptsov 2y ago
“Summing up: for the time being, if you run a social media company, or if you provide encrypted messaging services, which are accessible in France, and you’re based in the United States, get out of Europe.”
41.
by cheptsov 2y ago
Very interested in playing with their hardware and cloud. Also I wonder if it’s possible to try cloud without contacting their sales.
42.
by cheptsov 2y ago
Double that. The entire article reads to me as handpicked and manipulative.
43.
by cheptsov 2y ago
It’s called handpicking
44.
by cheptsov 2y ago
The author claims that everyone refers to Telegram as an encrypted messenger, but he only provides a single example to support that. I quickly checked Google News and couldn't find any media on the first page that did the same. It feel
45.
by cheptsov 2y ago
https://archive.ph/IybwU
46.
Mark Zuckerberg and Daniel Ek on why Europe should embrace open-source AI(economist.com)
2 points by cheptsov 2y ago|1 comments
47.
by cheptsov 2y ago
That’s on our list! We’re actually planning to expand our examples page with a lot more frameworks! Contributors are very welcome to our repo: https://github.com/dstackai/dstack/tree/master/examples/
48.
by cheptsov 2y ago
Oops, thanks for sharing this! Gonna update it shortly.
49.
by cheptsov 2y ago
Hi, I’m one of the contributors to dstack! Yes, you’re absolutely right—the title is a bit off. It should be "K8S for AI Teams Only." And yes, there is a lot of focus on LLMs specifically. If you have any other feedback, please sh
50.
Nous Hermes 3 and exploiting underspecified evaluations(interconnects.ai)
2 points by cheptsov 2y ago|0 comments
51.
by cheptsov 2y ago
I’m not a fan of Musk but it is so funny to see that many haters here.
52.
by cheptsov 2y ago
Profs?
53.
by cheptsov 2y ago
The one you disagree with but why do you think all people think the same way as you?
54.
by cheptsov 2y ago
I entirely agree. I’m quite sure Musk has a very strong stance on ethics, but it would be great to hear about it more clearly, and ideally not just through words, but through actual actions.
55.
by cheptsov 2y ago
Ahhhh huggingface_hub.utils._errors.HfHubHTTPError: 502 Server Error: Bad Gateway for url: https://huggingface.co/api/models/meta-llama/Meta-Llama-3.1-... Error: DownloadError
56.
by cheptsov 2y ago
They have shared the link to GitHub [1], HuggingFace repo [2], and the paper [3]: 1. https://github.com/aiola-lab/whisper-medusa 2. https://huggingface.co/aiola/whisper-medusa-v1 3. https:/&
57.
AiOla open-sources ultra-fast ‘multi-head’ speech recognition model(aiola.com)
71 points by cheptsov 2y ago|14 comments
58.
History and future of Infrastructure as Code (2023-05-13)(endoflineblog.com)
2 points by cheptsov 2y ago|0 comments
59.
Intel's Day of Reckoning Arrives: 15,000 Jobs Will Be Eliminated(extremetech.com)
57 points by cheptsov 2y ago|24 comments
60.
by cheptsov 2y ago
I wonder if the project makes it possible to run macOS inside a container on a macOS host?

More ›