How Hyperstition launched on X
Posted Nov 17, 2026 by @hyperstition_ccEngagement read Nov 19, 2026
Part 1: We now enter the era of hyper-composability. For the first time, it is economically feasible to serve individually fine tuned models at scale. And today, we open source it all. We can now serve heterogeneous batches of custom fine-tunes, without incurring LoRA’s per matrix bandwidth overhead. This allows us to economically serve this in large batches without incurring penalties for co-residing with default model users in the same batch. With Hypertune, we solve co-residence and are able to train co-resident adapters. Co-residence is the ability to compose independently trained fine-tunes into one model, without capability regression from adapter interference. Hypertune further matches LoRA in RL performance across 13 domains with 100x-500x fewer parameters. Serving heterogenous LoRA fine-tunes in one batch is prohibitively expensive. Each fine-tune variant needs its own low-rank adapter at every matrix and is only used for a few tokens, leading to many narrow matmuls and poor utilization. This is exacerbated for MoEs, where an adapter only engages when the token simultaneously selects a certain fine-tune and expert. This combinatorial blowup kills any remaining prospects of fast inference for fine-tuned / custom models. Hypertune introduces soft prefill, a trained KV adapter. Soft prefill can be trained using GRPO, SFT or any other first order method. As a consequence, Hypertune circumvents the LoRA overhead because soft prefills live entirely in the KV cache. This allows us to easily integrate it into existing serving stacks by prepending the soft prefills into paged KV blocks. Note: Context compression using cartridges requires copious amounts of synthetically generated self-study data for context distillation. Hypertune, instead, accommodates a panoply of gradient based methods for a wider range of objectives. Caveat: Comparisons with full-parameter GRPO are not required, because under RL, LoRA already matches it. LoRA Without Regret by @thinkymachines finds that LoRA performs equivalently to full fine-tuning for RL even at very small ranks, down to rank 1. The takeaway: Our gains indicate that substantial amounts of in-context learning capacity is left on the table by discrete prompts. This could be either because soft prompts reach conditioning states that no token sequence can express, or because their continuous parameterization makes good conditioning states far easier to find via gradient descent (even if a discrete equivalent exists). Companies and individuals can finally fine tune models on their private data, making models hyper-optimized to their needs, all without dramatically overpaying (75% cheaper than LoRA for inference). Here's the open sourced repo: https://t.co/359woJhSEK Part 2 coming out later this week.
- 353
- Likes
- 25
- Reposts
- 11
- Replies
- 9
- Quotes
- 364
- Bookmarks
How EPTA rated the launch
- Hook
- Editing / Creativity
- Copy
- Sentiment of launch
- Distribution strategy
How it compares
| Group | Launches | Median views | This one | Rank |
|---|---|---|---|---|
| AI infrastructure and chips launches | 127 | 187.1K | 0.29x | #93 / 127 |
| Product launchs | 513 | 125.8K | 0.43x | #379 / 513 |
| Accounts with 1K to 10K followers | 400 | 150.5K | 0.36x | #299 / 400 |
| Every launch | 963 | 147.7K | 0.37x | #721 / 963 |
What the numbers say
67 of every 10,000 viewers bookmarked it, above the directory median of 17. Bookmarks are the clearest sign a launch was worth coming back to.
The video runs 0:11, shorter than the median launch video at 0:53.
The post is 2804 characters, longer than the median launch post at 302.
It went out on a Tuesday. Launches posted on Tuesdays have a median of 156.7K views; Fridays lead at 235.5K.
Comparable launches
2026-11-19 · Source: EPTA launch directory, epta.co/directory. Free to use with attribution (CC BY 4.0).