331–340 of 362 posts
Earlier quoted context omitted.
I feel like once they make a mega successful movie of a book we can stop doing the secret handshake.
Especially when the movie is over forty years old.
This is more exciting than k3, IMO. Dsv4 models are extremely cheap to serve. Improving their capabilities has lots of downstream effects, as it becomes "good enough" for more and more tasks. DS was serving the pro version at extremely low prices for a long time, and they've had integrations with opencode & other providers, so they likely gathered a lot of data from real developers doing real tasks (on openrouter the…
we used to distill GLM 5.2 into Qwen 27B specialists with great success.
developing the workflows is tricky though. we had the advantage of a straightforward mapping function in mind: English -> SQL which was relatively easy to pipeline training for.
Earlier quoted context omitted.
So whoever is using DeepSeek V4 flash gets a massive upgrade without having to change any config.
That's an absolutely terrible idea for any non-trivial project consuming it.
If implementing non-trivial tasks on a preview model surely there is some anticipation of it not being the final stable release?
Unless you mean they should have appended a "preview" suffix on the endpoint for extra good measure?
Earlier quoted context omitted.
That's an absolutely terrible idea for any non-trivial project consuming it.
Wasn't it stated as being in "preview" prior to this? If implementing non-trivial tasks on a preview model surely there is some anticipation of it not being the final stable release? Unless you mean they should have appended a "preview" suffix on the endpoint for extra good measure?
Earlier quoted context omitted.
I feel like once they make a mega successful movie of a book we can stop doing the secret handshake.
The only movie I've ever enjoyed after reading the book was Jurassic Park. I've never seen the Dune movies or most other book-based movies, but honestly The Martian and Project Hail Mary were tempting. I'll probably surrender to one of those two someday.
Earlier quoted context omitted.
Works out of the box with OpenRouter for most models. Some providers are a bit flaky, but DeekSeek (provider) has been one of the most reliable for me, no problems hitting >99% CH.
The problem is when OpenRouter switches providers..
This is more exciting than k3, IMO. Dsv4 models are extremely cheap to serve. Improving their capabilities has lots of downstream effects, as it becomes "good enough" for more and more tasks. DS was serving the pro version at extremely low prices for a long time, and they've had integrations with opencode & other providers, so they likely gathered a lot of data from real developers doing real tasks (on openrouter the…
distilling K3 into DS4 flash will likely only be a good idea for specialists. the difference in model capacity is otherwise too large. we used to distill GLM 5.2 into Qwen 27B specialists with great success. developing the workflows is tricky though. we had the advantage of a straightforward mapping function in mind: English -> SQL which was relatively easy to pipeline training for.
How's their performance in English prose? We are currently searching for cost effective ways to keep story wikis up to date.
I am interested in this too, as I think it could be these models are overly optimized for coding. Let me know if you figure it out!
Earlier quoted context omitted.
distilling K3 into DS4 flash will likely only be a good idea for specialists. the difference in model capacity is otherwise too large. we used to distill GLM 5.2 into Qwen 27B specialists with great success. developing the workflows is tricky though. we had the advantage of a straightforward mapping function in mind: English -> SQL which was relatively easy to pipeline training for.
Why would you do this?
what we did was distill GLM 5.2 into a 27B model on SQL and then post-train it with RL afterward. the result outperformed even Fable on that one task. the distillation step is just good sense in this workflow, to bootstrap a smaller model to the utmost you can before actually doing RL.