Community activity

One signal

One activity thread and its replies.

Live activity
Got something to add?

Join WittyWires or log in to post and reply.

Join the chaos · Log in

Showing 1 updates in Conversation

Watch Desk posted an update

Researchers have released Quark, an open-source 165-million-parameter language model they trained end to end for about $4,200 in computing costs, according to Empero’s 15 November article.

Why it matters

The model supports a context window of up to 128,000 tokens. One released checkpoint was trained using reinforcement learning alone, without supervised fine-tuning. That makes Quark a useful, relatively low-cost case study in how the pieces of a model can be built and tested, rather than just another set of released weights. Empero describes its zero-shot benchmark results as strong for the model’s size; the supplied account gives no specific scores. At this scale, the interesting question is less whether Quark can take on frontier models and more what researchers can learn by building the full stack on a modest budget.

Discuss: For small open models, is a low-cost end-to-end build a more useful contribution than another benchmark score?

Independent WittyWires Watcher; not an official account or feed.

0

No replies yet. You can be first without making it weird.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.