Lemmy.World

169,988 readers
8,624 users here now

The World's Internet Frontpage Lemmy.World is a general-purpose Lemmy instance of various topics, for the entire world to use.

Be polite and follow the rules โš– https://legal.lemmy.world/tos

Get started

See the Getting Started Guide

Donations ๐Ÿ’—

If you would like to make a donation to support the cost of running this platform, please do so at the following donation URLs.

If you can, please use / switch to Ko-Fi, it has the lowest fees for us

Ko-Fi (Donate)

Bunq (Donate)

Open Collective backers and sponsors

Patreon

Liberapay patrons

GitHub Sponsors

Join the team ๐Ÿ˜Ž

Check out our team page to join

Questions / Issues

More Lemmy.World

Follow us for server news ๐Ÿ˜

Mastodon Follow

Chat ๐Ÿ—จ

Discord

Matrix

Alternative UIs

Monitoring / Stats ๐ŸŒ

Service Status ๐Ÿ”ฅ

https://status.lemmy.world

Mozilla HTTP Observatory Grade

Lemmy.World is part of the FediHosting Foundation

founded 2 years ago
ADMINS
ruud
Thekingoflorda
lwadmin
Rooki
AutoMod
kersploosh
MrKaplan
doccod
lwreport
bannanaente
listing: all - local
search
1
17
Qwen2.5: A Party of Foundation Models! (qwenlm.github.io)
submitted 3 months ago by brucethemoose to c/fosai
0 comments fedilink
 
 

cross-posted from: https://lemmy.world/post/19925986

https://huggingface.co/collections/Qwen/qwen25-66e81a666513e518adb90d9e

Qwen 2.5 0.5B, 1.5B, 3B, 7B, 14B, 32B, and 72B just came out, with some variants in some sizes just for math or coding, and base models too.

All Apache licensed, all 128K context, and the 128K seems legit (unlike Mistral).

And it's pretty sick, with a tokenizer that's more efficient than Mistral's or Cohere's and benchmark scores even better than llama 3.1 or mistral in similar sizes, especially with newer metrics like MMLU-Pro and GPQA.

I am running 34B locally, and it seems super smart!

As long as the benchmarks aren't straight up lies/trained, this is massive, and just made a whole bunch of models obsolete.

Get usable quants here:

GGUF: https://huggingface.co/bartowski?search_models=qwen2.5

EXL2: https://huggingface.co/models?sort=modified&search=exl2+qwen2.5

2
15
Qwen2.5: A Party of Foundation Models! (qwenlm.github.io)
submitted 3 months ago* (last edited 3 months ago) by brucethemoose to c/[email protected]
0 comments fedilink
 
 

https://huggingface.co/collections/Qwen/qwen25-66e81a666513e518adb90d9e

Qwen 2.5 0.5B, 1.5B, 3B, 7B, 14B, 32B, and 72B just came out, with some variants in some sizes just for math or coding, and base models too.

All Apache licensed, all 128K context, and the 128K seems legit (unlike Mistral).

And it's pretty sick, with a tokenizer that's more efficient than Mistral's or Cohere's and benchmark scores even better than llama 3.1 or mistral in similar sizes, especially with newer metrics like MMLU-Pro and GPQA.

I am running 34B locally, and it seems super smart!

As long as the benchmarks aren't straight up lies/trained, this is massive, and just made a whole bunch of models obsolete.

Get usable quants here:

GGUF: https://huggingface.co/bartowski?search_models=qwen2.5

EXL2: https://huggingface.co/models?sort=modified&search=exl2+qwen2.5

3
1
Qwen2.5: A Party of Foundation Models (qwenlm.github.io)
 
 

Comments

view more: next โ€บ