coding models have gotten much better at ui generation, but still struggle with stylistic and domain-specific requirements.
we worked with @gg_lemonade to post-train hy3 to more consistently build roblox uis that follow their visual styles.
here’s how: castform.com/blog/lemonade
@castformai now supports post-training vision models 👁️
you can train on images, diagrams, and screenshots as input, beyond just text.
quick thread on what this unlocks 🧵
1/
you can now train on your @harborframework tasks with @castformai .
point at a taskset, declare your harness + sandbox, and you're good to go. ~20 lines.
sft's great when you have clear ground truths and 1000s of labelled examples. where you should consider using it:
classification and structured extraction: examples include routing, intent classification, field extraction. there is a clear correct answer & you likely have 1000s of labeled samples. finetuning a small model will help you match a much larger one, at a fraction of the cost & latency.
distillation from a larger model: generate traces from a frontier model on your task, then train a smaller one on them. for simpler tasks, you can match quality and get a model that’s cheaper to serve and fast enough to sit in a hot path.
warm-start for rl: reinforcement learning needs the model to get to a non-zero score to be able to reinforce useful behaviors. for more complex tasks, base models might not be able to get there. sft on expert traces/ground truths can help move it to the right part of the distribution first, so rl can work its magic.
castform now supports supervised finetuning (sft)
our goal’s to give developers everything necessary to post-train models for their tasks. rl is all the rage these days (& castform makes it super easy to use), but sft's one of most compute-efficient ways to teach a model exactly how you want it to behave.
we sft'd a 4b model to do pii redaction. 3,000 examples in, it's outperforming gpt-5.6-sol and at much lower cost. 👇
"we don't have training data" and "post-training needs infra we don't have" are the two primary reasons teams pass on rl post-training.
castform + neon kills both blockers for training fast, accurate, cheap search agents.
turn your existing neon data to training data and then post-train a 4b model that beats frontier llms on price and efficiency.
neon.com/blog/how-castform-n…
A 4B open-source model post-trained with @castformai matched GPT-5.6 Sol on search accuracy at ~100x lower cost
The corpus backend for the whole pipeline: Neon + Lakebase Search
frontier models are excellent english speakers.
turns out, that can make them bad english teachers.
we worked with @ELSA_Speak to post-train a better, smaller, faster model on learner conversations - making responses simpler, more helpful, and half the cost and latency.
here’s how:
girish @googrish Jun 23 Article designing dev onboarding for an agent-first world tl;dr we’re building an rl post-training platform, targetting developers who have never post-trained a model. we figured out the training, but the open problem was onboarding. we tested different
lots of talk about agi, asi, rsi but ask any frontier LLM to roll a die and it will almost always say "4." claude, gpt, kimi - doesn't matter, 4.4.4.4.
so here's how i post-trained a model to reliably roll a die (i.e. each number ~1/6th of the time) & why it's a nice sandbox for one of the most interesting problems in rl i.e. getting a model to actually explore instead of just following strategies it already knows 🧵
with the events around fable, it’s clear that companies & developers need to own their models.
the ability to post-train & rl models must become a more broadly accessible skill.
rl fine tuning sounds like rocket science, but it really isn’t.
so @Thariq_q (not the claude code guy :D) made a video that explains it with as little jargon as possible👇
one of the most exciting surprises from the launch has been seeing just how diverse the developer use cases are for custom models. human creativity is really unmatched :)
we really want to help bring more of these ideas to life.
so, we’re giving away $1,000 in free credits to folks with compelling ideas they want to train.
just reply to the post with your idea, or email us at castie[at]castform[dot]com
castform is in open beta!
our goal’s to enable any developer post-train their own llms. in the world of rapidly rising llm costs & providers guarding capabilities, we believe the ability to shape model behavior shouldn’t be a privilege.
this release today is our small step towards fixing that.
$50 in free credit for new users. hope you’ll give it a try 🙂
“don’t train your own model” is common ai advice. it's wrong. your token bill's the proof.
today, we’re excited to launch castform into open preview. castform is the easiest way for you to train your own model, on your own data.
open-weights models are performant and much cheaper. when trained on your task & proprietary data, they beat closed models. the thing standing between you and that was weeks of plumbing & years of ml expertise.
with castform, model training is as simple as prompt engineering. @castformai
bring your agent traces or raw corpora. castform turns it into training data, picks the right algorithmic recipes, manages gpus, and gives you an ide to watch and chat with your model as it learns.
see what you can build with castform👇