covi

7 ms·

31.
SkyServe: 50% Cheaper AI Serving on Any Cloud with High Availability(blog.skypilot.co)
5 points by covi 3y ago|0 comments
32.
Tigris: Globally Distributed S3-Compatible Object Storage(tigrisdata.com)
6 points by covi 3y ago|0 comments
33.
Scaling Mixtral LLM Serving Across Clouds(blog.skypilot.co)
2 points by covi 3y ago|0 comments
34.
by covi 3y ago
Now just need a Waymo invite code :)
35.
by covi 3y ago
Cloud deployment docs: https://docs.mistral.ai/cloud-deployment/skypilot/
36.
Covariant: Moving from on-prem to the cloud for 4x faster AI development(blog.skypilot.co)
4 points by covi 3y ago|1 comments
37.
by covi 3y ago
https://www.forbes.com/sites/alexkonrad/2023/07/13/ai-startu... > Its revenue run rate has spiked this year and now sits at around $30 million to $50 million, three sources said — with one noting
38.
ML experiments in the cloud with SkyPilot and DVC(alex000kim.com)
2 points by covi 3y ago|0 comments
39.
by covi 3y ago
Part of the Vicuna team wrote a guide on finetuning Llama2: https://blog.skypilot.co/finetuning-llama2-operational-guide...
40.
by covi 3y ago
Related ongoing thread: https://news.ycombinator.com/item?id=36975245
41.
by covi 3y ago
You could finetune then add on a retrieval step, which has the advantage of citing sources. Jury's probably still out on which, or which combination of, methods work best. Likely use case and data size-dependent.
42.
Cookbook: Finetuning Llama 2 in your own cloud environment, privately(blog.skypilot.co)
119 points by covi 3y ago|13 comments
43.
Run Llama2 in your cloud, completely privately(github.com)
3 points by covi 3y ago|1 comments
44.
by covi 3y ago
Meta's infra and system teams have been on a roll. Feels like Google in the 2000s. Check out quite a few interesting papers from them in OSDI 2023 (top systems conference, co-located with ATC where the OP URL links to): https:/&#
45.
The Production Environment at Google, from the Viewpoint of an SRE(sre.google)
2 points by covi 3y ago|0 comments
46.
UC Berkeley's SkyPilot project: unprecedented GPU availability across new clouds(blog.skypilot.co)
5 points by covi 3y ago|0 comments
47.
by covi 3y ago
Twitter announcement: https://twitter.com/lmsysorg/status/1646618380114468864
48.
UC Berkeley's open-source Vicuna LLM chatbot released new improved model weights(huggingface.co)
26 points by covi 3y ago|3 comments
49.
by covi 4y ago
Kudos to Databricks! Anyone has insights into benchmark & real-world quality? From https://huggingface.co/databricks/dolly-v2-12b#benchmark-met... , it seems like dolly-v2-12b's benchmark results are actually s
50.
by covi 4y ago
This is awesome, and glad to see that SkyPilot is useful in distributing Tabby on any cloud! https://github.com/TabbyML/tabby/blob/main/deployment/skypil...
51.
by covi 4y ago
From the post: > The training was done with PyTorch FSDP on 8 A100 GPUs in one day. > We employ SkyPilot managed spot to reduce the cost by leveraging the cheaper spot instances with auto-recovery for preemptions and auto zone switch.
52.
Show HN: Run LLaMA LLM chatbots on any cloud with one click(github.com)
4 points by covi 4y ago|1 comments
53.
Run LLaMA LLM chatbots on any cloud with one click(blog.skypilot.co)
12 points by covi 4y ago|0 comments
54.
by covi 4y ago
We don't anticipate this will happen for a few reasons. When the usage of SkyPilot (or a SkyPilot-like "intercloud broker" system) is small, it probably doesn't warrant the dominant clouds' attention. When the usage
55.
by covi 4y ago
SkyPilot devs here, happy to answer any questions. GitHub repo (Apache 2 license): https://github.com/skypilot-org/skypilot Getting started is easy: $ pip install "skypilot[aws,gcp,azure]" # Pick your clouds
56.
by covi 4y ago
One big reason people use multiple regions or clouds: higher resource availability. Allocating scarce resources can be very hard on the cloud. These are things like high-end GPUs (both on-demand and spot; the latter being much harder to get
57.
by covi 4y ago
+1. We've heard from some heavy users that Cloudfare R2 is saving them $$$ on egress costs: https://www.cloudflare.com/products/r2/ As outlined in the position paper (linked by another commenter) we believe s
58.
by covi 4y ago
(SkyPilot dev) Can't agree more on autostop / autodown. Subjective to measure but save our own costs massively.
59.
by covi 4y ago
I'm one of the creators of SkyPilot. Thanks for the thoughtful questions and let me try to take a stab: SkyPilot is not just for multi clouds. It's useful for all of these scenarios: - using a single region of one cloud - using mu
60.
by covi 4y ago
Indeed! Having lower-cost GPU clouds in the "Sky" is on our immediate roadmap: https://github.com/skypilot-org/skypilot/blob/master/ROADMAP... In fact, as we speak we're working with folks

More ›