Tianyu Fu

Ph.D. student at NICS-EFC Lab, Tsinghua University; visiting student at InfiniAI Lab, CMU.

[Tianyu Fu @fuvty123]

Check out our new paper on multi-LLM communication!
Welcome to any questions or discussions 🙌

[机器之心 JIQIZHIXIN @jiqizhixin]

Wow, language models can talk without words.

A new framework, Cache-to-Cache (C2C), lets multiple LLMs communicate directly through their KV-caches instead of text, transferring deep semantics without token-by-token generation.

It fuses cache representations via a neural projector and gating mechanism for efficient inter-model exchange.

The payoff: up to 10% higher accuracy, 3–5% gains over text-based communication, and 2× faster responses.

Cache-to-Cache: Direct Semantic Communication Between Large Language Models

Code: https://t.co/swjJm2gssr
Project: https://t.co/b21mjmPMXK
Paper: https://t.co/BfwOpGldNA

Our report: https://t.co/xj6FCALfr1

📬 #PapersAccepted by Jiqizhixin

[Tianyu Fu @fuvty123]

👀

[Benhao Huang @huskydogewoof]

𝐇𝐨𝐰 𝐝𝐨 𝐰𝐞 𝐠𝐞𝐭 𝐟𝐫𝐨𝐦 𝐚 𝐬𝐭𝐚𝐧𝐝𝐚𝐫𝐝 𝐟𝐞𝐞𝐝𝐟𝐨𝐫𝐰𝐚𝐫𝐝 𝐦𝐨𝐝𝐞𝐥 𝐭𝐨 𝐚 𝐜𝐚𝐩𝐚𝐛𝐥𝐞 𝐢𝐭𝐞𝐫𝐚𝐭𝐢𝐯𝐞 𝐦𝐨𝐝𝐞𝐥?

On Sudoku, we traced the exact path of unlocking neural attractors:

- Feedforward → 2.6%
- Weight-tying → 32.6%
- Online Training → 74.7%
- Hierarchy → 76.5%
- Adaptive Compute → 84.8%

Each jump wasn't just a trick. It was a choice about how to shape the attractor landscape.

Here is what we learned: 🧵👇

#ICML2026

[Tianyu Fu @fuvty123]

Interesting work. Congratulations on Benhao :)

[Benhao Huang @huskydogewoof]

🌀 Introducing 𝐄𝐪𝐮𝐢𝐥𝐢𝐛𝐫𝐢𝐮𝐦 𝐑𝐞𝐚𝐬𝐨𝐧𝐞𝐫𝐬 (𝐄𝐪𝐑) !

Feedforward models and weight-tied models behave very differently on hard reasoning generalization.

EqR pushes this difference to the extreme by learning 𝐭𝐚𝐬𝐤-𝐜𝐨𝐧𝐝𝐢𝐭𝐢𝐨𝐧𝐞𝐝 𝐧𝐞𝐮𝐫𝐚𝐥 𝐚𝐭𝐭𝐫𝐚𝐜𝐭𝐨𝐫𝐬 .

• Sudoku-Extreme: 99.8%
• Maze: 93%

#ICML2026

[Tianyu Fu @fuvty123]

Amazing work! Congratulations to @yi_xin_dong !

[Yixin Dong @yi_xin_dong]

Introducing XGrammar-2: structured generation for complex agent harnesses.

Strict tool-calling formats. Built-in DeepSeek-V4 and Qwen-3.6 support. Up to 80x speedup over XGrammar. Ready-to-use integrations with vLLM, SGLang, TensorRT-LLM, and more! ⚡

From Claude Code to OpenClaw, agents are defining more complex harnesses. XGrammar-2 ensures LLMs always interact with them in the right way.

Built in collaboration with DeepSeek, Databricks, and leading frontier AI labs to bring XGrammar-2 into latest models and products.

🧩 Structural Tag: one unified abstraction to describe any format your agent needs
🚀 Scales to 500+ strictly typed tools for complex agent harnesses
🌐 Native APIs in Python, C++, Rust, and JS, running everywhere from cloud to edge
🛠️ Integrated with vLLM, SGLang, TensorRT-LLM, and more

Excited to see what agent builders create with it!

Blog: https://t.co/N0Tbl588BH
GitHub: https://t.co/lo4yScuI2f

[Tianyu Fu @fuvty123]

Awesome looped transformers list from @huskydogewoof . Such a timely addition to the ever-growing looped transformers community!

https://t.co/hqYKte6d84

https://t.co/FyLqpCHpgD

[Benhao Huang @huskydogewoof]

Introducing 🔁 Awesome-Loop-Models: a curated repo for keeping up with loop models!

Whether you are just entering the field or have been exploring loop models for a while, this repo is built to serve as an actively updated map for mechanism analysis, architecture and algorithm design, applications, and related directions.

🧵 [1/n]

[Tianyu Fu @fuvty123]

@sukjun_hwang @_albertgu @fluorane Saw this poster. A really nice one!

[Tianyu Fu @fuvty123]

@LIT_workshop Author of Think-at-Hard here 🙋 I don’t use X much, so didn’t get tagged, but I’d be happy to chat more about the work 😊 Thanks so much for hosting the workshop!

[Tianyu Fu @fuvty123]

@huskydogewoof Thank you so much for sharing~ Looking forward to more improvements in the field of looped transformers and implicit thinking!

[Tianyu Fu @fuvty123]

Thank @LIT_workshop for sharing Think-at-Hard. Looking forward to exploring more about looped transformers at ICLR.

[Latent & Implicit Thinking Workshop @LIT_workshop]

🏆 LIT Workshop @ ICLR 2026 — Community Choice Award!
Vote for your favorite paper from our Best Paper finalists 👇
Details on each paper in the thread 🧵

[Tianyu Fu @fuvty123]

@moltbook Wait until moltbots can talk without words.
The method is already there, wonder when will moltbots find out: https://t.co/p0TVswKvpE

[Tianyu Fu @fuvty123]

Clawbots @openclaw are everywhere on @moltbook .
Now imagine if they could 💬 talk without words 😶‍🌫️

They can! 🤯

Cache-to-Cache (ICLR’26) lets LLMs communicate directly with KV, beyond text.

Webpage: https://t.co/p0TVswKvpE

#cache2cache #Clawbot #moltbook

[Tianyu Fu @fuvty123]

Congratulations to @RJ_Sadhukhan and @InfiniAILab on the interesting exploration of embedding modules! It feels like new shifts in FFN architectures are on the move 🏃‍♂️

[Infini-AI-Lab @InfiniAILab]

Lookup memories are having a moment 😄
The whale 🐋 #deepseek dropped engram… and we dropped up-projections from our FFNs…perfect timing 😅

🥳 Introducing STEM: Scaling Transformers with Embedding Modules 🌱
A scalable way to boost parametric memory with extra perks:
✅ Stable training even at extreme sparsity
✅ Better quality for fewer training FLOPs (knowledge + reasoning + long-context gains)
✅ Efficient inference: ~33% FFN params removed + CPU offload & async prefetch
✅ More interpretable → seamless knowledge editing 🔧🧠
Looking forward to DeepSeek v4… feels like we’ve only scratched the surface of embedding-lookup scaling 👀

📄Paper: https://t.co/ecyOtgb6sv
🌐 Website: https://t.co/RXquIha62p
🔗 GitHub: https://t.co/5K05Lm4ncE

[Tianyu Fu @fuvty123]

@tyao923 Very interesting work! In our previous work, Think-at-Hard, we also explored weighted summation over token embeddings with sampling probabilities, following Soft Thinking. What are your thoughts on sample then aggregate versus weighted aggregate?

[Tianyu Fu @fuvty123]

Congratulations on the amazing work! We also worked on token-level routing in R2R (https://t.co/JAGasLo4vs). It would be great if the framework could extend the support to token-level routing as well 🙌

[Jiaxuan You @youjiaxuan]

Huge congrats to the LLMRouter team for hitting 1,100 GitHub stars in just one week! ⭐ The excitement was way beyond the team's expectations.

Thanks to community feedback, the LLMRouter team has already shipped major updates:

🆕 What's New:

🔧 Unified Configs: Seamlessly route across mixed backends—Cloud (OpenAI, Anthropic, Gemini, NVIDIA) and Local (vLLM).

🎥 Multimodal Support: Now handling Video/Image + Text routing across Geometry3K, MathVista & Charades-Ego.

💻 Code: https://t.co/RYrGZnTD8x
📄 Project Page: https://t.co/b2SYselcL9

[Tianyu Fu @fuvty123]

@youjiaxuan @thinkymachines Sounds very useful! Plan to try it out.

[Tianyu Fu @fuvty123]

@yuz9yuz @patpcj @jclin808 Congratulations on the amazing survey!

[Tianyu Fu @fuvty123]

@wenhaocha1 Good points 👍

[Tianyu Fu @fuvty123]

Congratulations to Jintao on the amazing work to speed up diffusion models!

[Jintao Zhang @Jintao_Zhang_]

TurboDiffusion: 100–205× faster video generation on a single RTX 5090 🚀

Only takes 1.8s to generate a high-quality 5-second video.

The key to both high speed and high quality?
😍SageAttention + Sparse-Linear Attention (SLA) + rCM

Github: https://t.co/vT3nfax8H9

Technical Report: https://t.co/LEgLyhdPXh

[Tianyu Fu @fuvty123]

@KTL_XAI Thanks for the great question.
Figure 1 compares standard and TaH models, which have different weights because they are trained separately. The “correct→wrong” means the standard model gets the answer right, but the TaH model gets it wrong with the oracle iteration policy.

[Tianyu Fu @fuvty123]

Thanks @_akhaliq! We warmly welcome any discussions on HuggingFace Daily Paper :)
https://t.co/p4PiNmD4Bm

[AK @_akhaliq]

Think-at-Hard

Selective Latent Iterations to Improve Reasoning Language Models

[Tianyu Fu @fuvty123]

@TheTuringPost We are also exploring latent communication between LLMs with a paper called "cache-to-cache". It is really nice to see the multi-LLM community growing so fast!

[Tianyu Fu @fuvty123]

@Starc_Institute Thank you for the wonderful breakdown and for sharing our work! Open to any questions or discussions :)