🦍 Introducing Memorilla: memory as a modality for LLMs
Memorilla turns everything a frozen LLM needs to remember into just 16 memory embeddings
⚡ It uses up to 4,212× fewer tokens and is 27× faster than full context inference while more accurate
🎯 Pushes the Pareto frontier over RAG and agentic memory in personalization, long-document and knowledge QA
A thread 🧵:
Reframing cartridge training as a self-play game cuts the training steps needed by more than half, just by asking better questions.
This post continues our series on cartridges, compressed KV caches of long documents. Here we focus on self-study, the method used to train them. Self-study teaches a model reading the cartridge to answer the way it would with the full document in context. Currently, its questions come from fixed, generic prompts, so many of them ask about things the cartridge already knows.
Recent KV cache compression research has focused on small models with dense attention. We scaled these techniques to GLM 5.3, a frontier sparse attention model.
We adapted KVzip and Attention Matching, reducing KV cache size by 80% while retaining over 90% of full-context accuracy on the QuALITY benchmark.
Readers added context they thought people might want to know
The video is an animation by artist R.J. (@RJ16848519) from his Screen Time series, shared here without credit.
x.com/rektober/statu…
x.com/connoissurreal…
x.com/kikanicolela/s…
there are many many people who studied physics and came to the conclusion that working on machine learning would probably create more scientific progress than studying physics including many of the original chatgpt crew like john schulman, liam fedus, amodei & jared kaplan at ant
Introducing Ramp Accounting Bench.
We partnered with accounting professionals to create 137 tasks and grading rubrics grounded in everyday accounting workflows.
Even with three attempts, the best model achieved only 21% accuracy. Reliable agentic accounting remains an open challenge.
Ben Geist retweeted
last week I hit my 3 years at @modal where i've been leading design across product + brand and slowly building out our team. we've been investing more in brand as the company has grown and i'm excited to announce that @studiojud has joined us as our Director of Brand.
we worked together last fall as Studio Jud's first client, and i've spent the last year slowly convincing him to join 💚
...and we’re hiring more designers: modal.jobs
AI spend is measured in tokens, model calls, and dollars. However, none of these fields describe the actual work being done.
At Ramp, we built a semantic layer that attributes agent spend to objectives and outcomes. This allowed us to go from monitoring AI spend to understanding AI ROI.
Here's how we built it 🧵
Ben Geist retweeted
Monitor and control your AI spend on every provider on router.com. Our early users save 40% on average.
Every week, the price-intelligence-latency frontier shifts, and we expect this trend to continue. Tradeoffs between latency, reasoning, cost, service tier, open source and closed source models are shifting constantly.
Router sends every request to the model that's actually best for the task and helps you control what tokens you buy. We benchmark it against real work: ~40% lower cost for the same outputs.
Today we're opening it to everyone.
Two lines of code or just change your base URL. No @tryramp account needed.
Free through 2026, first $26 on us.
Get an API key today at router.com
Ben Geist retweeted
honored to be one of the alpha testers of @tryramp PorTAL, and thank you @b_geist for resolving the issues at lighting speed!
was able to get the full PorTAL process running within a virtual machine, grab repo here: github.com/NickyHeC/smol-por…
built on: github.com/smol-machines/smo…
We’re open-sourcing PorTAL, our framework for shared task representations and cross model LoRA adaptation.
It now spans from hybrid attention models to multimodal systems including Gemma 4 E2B, Mistral 7B & @thinkymachines' Inkling.
Code: ramp-public/portallib
Models: @huggingface /RampPublic
Ben Geist retweeted
Today we’re launching Ramp Router.
3 years ago, we built an internal LLM router at @tryramp that powers AI products for 70,000 customers.
Back then it was mostly about saving money.
Now it feels obvious: the best model changes constantly. GPT, Claude, Gemini, Grok, Qwen, DeepSeek, Kimi, GLM - prices and capabilities move every week.
So we’re opening up access to everyone.
One OpenAI-compatible endpoint. The right model for every request. Lower cost without rewriting your app.
Reserve access to use it.
Introducing PorTAL: Portable Task Adapters for LLMs. A novel recipe to cheaply port fine-tuning between models. It matches per task LoRA accuracy at half the cost, lowering the switching overhead of adapting tasks across LLMs.
At Ramp, every new model release used to mean retraining our fine-tunes from scratch. PorTAL learns the task once, then efficiently refits it onto any new base model, even across model families.
Introducing PorTAL: Portable Task Adapters for LLMs. A novel recipe to cheaply port fine-tuning between models. It matches per task LoRA accuracy at half the cost, lowering the switching overhead of adapting tasks across LLMs.
At Ramp, every new model release used to mean retraining our fine-tunes from scratch. PorTAL learns the task once, then efficiently refits it onto any new base model, even across model families.


















