@RomAI__
Account based inFrance
About this account
- Account based in
- France
- Connected via
- France App Store
Account-level information from X, not a live location or the device used for a specific post.
Turning scattered compute into faster, open intelligence. @reliquary_ai
Joined May 2020
- Tweets74
- Following254
- Followers125
- Likes357
romAI retweeted
GLM 5.3’s biggest leap came from post-training alone.
“GLM 5.3 was built entirely by post-training the previous model. For me, that was a wake-up call.”
“Post-training is far more GPU-efficient than pre-training from scratch. Plenty of companies can’t afford to train a model from scratch, but they have proprietary data. They can post-train themselves and beat some of the closed models.”
“There’s still so much juice left to squeeze from post-training. We compared GLM 5.3’s cyber capabilities with 5.2, and it’s a night and day difference. That’s all just post-training.”
This makes decentralized RL a lot more scalable.
If samplers only need to stay within a bounded mismatch, far more GPUs and precisions can feed the same training run.
That’s what we’re building at @reliquary_ai, permissionless. Come see it run.
Training large models with RL typically runs on a fixed homogeneous cluster of the latest accelerators for numerical stability and faster throughput. This throws away inference optimization work done for other vendors and generations, limits the capacity you can utilize, and is unnecessarily expensive.
So we trained a model on agentic coding tasks using fully asynchronous RL and mixed-precision, mixed-accelerator samplers. With off-policy correction, training stayed stable through convergence, on par with homogeneous samplers, at a lower cost.
If you have a GPU doing nothing, read this.
Every idle GPU could be doing post-training.
Post-training doesn’t need a datacenter. It needs GPUs that are free.
A centralized lab can only use the GPUs it owns or rents.
Reliquary can use any GPU in the world.
That’s why we build on Bittensor.
Here’s the key. Inference means low latency and no interruptions: an answer can’t stop halfway.
Post-training doesn’t need either. Nobody is waiting on a rollout. If a GPU drops mid-task, nothing breaks.
So any GPU can contribute. Idle at night, free between experiments. Join for an hour, leave, come back. The work still counts.
A centralized lab pays for its GPUs 24/7, busy or not.
We only pay for work that’s delivered and accepted.
And every GPU owner can know what they’ll earn before they start.
Our API now publishes, live, what each task pays per million tokens. Check your card’s throughput, and you have your number.
reliqua.ai/api/sft/pay
romAI retweeted
People are laughing at data companies making $1B a year and call it NGMI. Entirely wrong.
As RSI becomes more accessible, the bottleneck shifts from "a smart math genius knowing training recipes" to "access to high-quality data". Data companies create the most value in this new world.
AI is the biggest math genius. We don't need more math geniuses. We need the most unique, best curated data with a strong point of view. Whoever owns this can either sell to labs, or train their own model that beat the labs.
Just accept the new reality, guys.
romAI retweeted
A look at our research infra.
High-quality training data requires intuition built by training models and studying their failures. As tasks get more complex, this matters more.
Creating data without training on it is like building a product you never use. Excited to share more.
Working on RL / post-training?
We just opened early access to Reliquary's batch API: low-cost training data generation at scale.
Bring your environment and tell us what you're training.
DM me to try it!
reliqua.ai
romAI retweeted
Reliquary is building post-training on demand.
The network behind our own models is opening up to Bittensor subnet teams.
First up: Batch generation. Training data built around your model and tasks.
Early access:
app.reliqua.ai/early-access
Felt exactly the same. After those conversations I've never been more bullish: frontier models built end to end on Bittensor are coming. Great meeting you guys, let's make it happen.
Now that we’ve had a moment to process everything from @ExploitSummit, the energy and motivation to build are stronger than ever after all the conversations we had.
Exchanging ideas with decentralized training teams like @MacrocosmosAI or @reliquary_ai and hearing @jondurbin talk about Parallax made one thing clear: Bittensor has the talent and energy to build models that can compete with SOTA. That's exactly what we’re working toward with @teutonic_tony .
Really glad I got to be there with @DendriteHQ
romAI retweeted
Finally, a RL simulation environment for port logistics 🚢 (v1)
A lot of you said this should exist. So built it using real world port data. 👀
romAI retweeted
A week ago we added a new task to Subnet 81: batch generation. You give it a dataset and a model, miners generate the completions.
Since then, the network has generated nearly 3B tokens with Qwen3.8-27B, with outputs spot-checked against the model's weights.
We're using them to distill into Teutonic, a model from Bittensor.
Next: making this available to everyone. More on Tuesday.
Met more builders in two days at the @ExploitSummit than in the past year.
Came home with more energy and a longer to-do list
Thanks Const
Next run is coming soon, with bigger environments and a harder target
What the Reliquary team has invented for a minable objective is truly one of the most compelling structures i have seen for a scalable post training mechanism on Bittensor.
In the same way that conjectures 66 searches math proofs, Reliquary has miners build compute fleets which search for rollouts on a target model which partition the reward landscape.
You spin up a miner, download the network model and scale your inference engine + innovate on your search technique to uncover negative samples from the environments.
Those samples become optimal inputs to the training process for the network model.
Like how Bitcoin miners bleed energy for a hash. Reliquary miners bleed energy for a rollout. It’s a search problem at heart which is where Bittensor’s abstraction over the search for search algorithms really shines.
Reliquary was created by two young engineers with no credentials and no allegiances to anything except a vision for decentralized intelligence.
Bravo Rom and Grizz
romAI retweeted
@RomAI__ of @reliquary_ai announced early access for Bittensor subnet teams at Exploit Summit. With cofounder @0xgrizzz.
With Reliquary-4B released, the next step is post-training on real customer tasks, tailoring the work to what each team's model needs to do.
romAI retweeted
Post-training is where frontier models get sharper - and where compute gets expensive.
At Exploit, @0xgrizzz + @RomAI__ of @reliquary_ai show how SN81 turns reinforcement learning into an open market: miners hunt for training signal, prove the work + compete to improve the next checkpoint: luma.com/exploitsummit26
romAI retweeted
We just shipped a major upgrade to Reliquary's validation system.
Validators no longer need to load the full model. They stream it layer by layer, so a single H100 can now verify a frontier MoE, bit-identical to running it whole.
Bigger models are coming to Reliquary.
romAI retweeted
Introducing Reliquary-4B.
A 4B math & code model trained with reinforcement learning. Anyone could join the network and contribute rollouts.
Independent miners chose the prompts and generated the rollouts. The protocol verified them and trained the model.
Here’s the model and the research behind it.
🤖 Made with AI
