Amr retweeted
You can use Claude Fable 5.2 to make MSAs. The MSAs have clear coevolution signal. You can give those MSAs to AlphaFold for structure prediction 🤠 (1/5).
Amr retweeted
🚀 Introducing DeepInfer, our in-house inference framework, now fully open source!
Powering DeepSeek-V4.1-Flash at scale with high throughput and low latency. We’re sharing our production stack to help the community lower compute costs, reduce energy use, and make AI more affordable.
🔗 github.com/deepseek-ai/DeepI…
1/4
Amr retweeted
I am excited to announce that @Coinbase has just released its MPC engine as open source github.com/coinbase/cb-mpc. The library provides two-party and multiparty signing for ECDSA and Schnorr/EdDSA, as well tools for DKG, backup and more. 1/6
Amr retweeted
1/ WE TOOK TERRITORY by issuing the first bank-issued #stablecoin on a permissionless blockchain & it's not what you think. 🧵 below. The real impact is on #tradfi--yes, #crypto took regulatory territory, but #tradfi is the real story in what @custodiabank did w/ @Vantage_Bank.
Just a few weeks left in Q1—if you haven’t tried LLMs yet, it’s like refusing to use calculators because you’re loyal to your abacus. And if you’re worried AI might take your job, remember: robots can’t replace your killer sense of humor. Or can they?
Today, we release QwQ-32B, our new reasoning model with only 32 billion parameters that rivals cutting-edge reasoning model, e.g., DeepSeek-R1.
Blog: qwenlm.github.io/blog/qwq-32…
HF: huggingface.co/Qwen/QwQ-32B
ModelScope: modelscope.cn/models/Qwen/Qw…
Demo: huggingface.co/spaces/Qwen/Q…
Qwen Chat: chat.qwen.ai
This time, we investigate recipes for scaling RL and have achieved some impressive results based on our Qwen2.5-32B. We find that RL training con continuously improve the performance especially in math and coding, and we observe that the continous scaling of RL can help a medium-size model achieve competitieve performance against gigantic MoE model. Feel free to chat with our new models and provide us feedback!
1
97
Amr retweeted
We open sourced our decentralized training framework prime 🫡
For efficient, globally distributed training of AI models over the internet.
Features:
- ElasticDeviceMesh for Fault-Tolerant Training
- Asynchronous Distributed Checkpointing
- Live Checkpoint Recovery
- Custom Int8 All-Reduce Kernel
- Maximizing Bandwidth Utilization
- PyTorch FSDP2 / DTensor ZeRO-3 Implementation
- CPU Off-Loading
You’re unable to view this Post because this account owner limits who can view their Posts. Learn more
Amr retweeted
This is really a 'WOW' paper. 🤯
Claims that MatMul operations can be completely eliminated from LLMs while maintaining strong performance at billion-parameter scales and by utilizing an optimized kernel during inference, their model’s memory consumption can be reduced by more than 10× compared to unoptimized models. 🤯
'Scalable MatMul-free Language Modeling'
Concludes that it is possible to create the first scalable MatMul-free LLM that achieves performance on par with state-of-the-art Transformers at billion-parameter scales.
📌 The proposed MatMul-free LLM replaces MatMul operations in dense layers with ternary accumulations using weights constrained to {-1, 0, +1}. This reduces computational cost and memory utilization while preserving network expressiveness.
📌 To remove MatMul from self-attention, the Gated Recurrent Unit (GRU) is optimized to rely solely on element-wise products, creating the MatMul-free Linear GRU (MLGRU) token mixer. The MLGRU simplifies the GRU by removing hidden-state related weights, enabling parallel computation, and replacing remaining weights with ternary matrices.
📌 For MatMul-free channel mixing, the Gated Linear Unit (GLU) is adapted to use BitLinear layers with ternary weights, eliminating expensive MatMuls while maintaining effectiveness in mixing information across channels.
📌 The paper introduces a hardware-efficient fused BitLinear layer that optimizes RMSNorm and BitLinear operations. By fusing these operations and utilizing shared memory, training speed improves by 25.6% and memory consumption reduces by 61% over an unoptimized baseline.
📌 Experimental results show that the MatMul-free LLM achieves competitive performance compared to Transformer++ baselines on downstream tasks, with the performance gap narrowing as model size increases. The scaling law projections suggest MatMul-free LLM can outperform Transformer++ in efficiency and potentially in loss when scaled up.
📌 A custom FPGA accelerator is built to exploit the lightweight operations of the MatMul-free LLM. The accelerator processes billion-parameter scale models at 13W beyond human-readable throughput, demonstrating the potential for brain-like efficiency in future lightweight LLMs.
Amr retweeted
i never believe recorded demos so I reached out to the @cognition_labs team for early access to try for myself and got it!
will be sharing my unfiltered opinions on #devin here.
🧵🧵 1/n
Today we're excited to introduce Devin, the first AI software engineer.
Devin is the new state-of-the-art on the SWE-Bench coding benchmark, has successfully passed practical engineering interviews from leading AI companies, and has even completed real jobs on Upwork.
Devin is an autonomous agent that solves engineering tasks through the use of its own shell, code editor, and web browser.
When evaluated on the SWE-Bench benchmark, which asks an AI to resolve GitHub issues found in real-world open-source projects, Devin correctly resolves 13.86% of the issues unassisted, far exceeding the previous state-of-the-art model performance of 1.96% unassisted and 4.80% assisted.
Check out what Devin can do in the thread below.
Amr retweeted
1/ Last week, we caught up with @PrestonPysh to discuss @CustodiaBank's segregated #Bitcoin custody service. Check it out on your preferred podcast app to learn more about the latest developments for Custodia. Here’s a 🧵about what we covered👇
youtube.com/watch?v=u8UpiYJq…
Amr retweeted
1/ BIG NEWS!! Custodia’s #Bitcoin custody platform recently went live. Custodia is a *BANK* built by Bitcoiners & we offer segregated (not omnibus) custody accounts on our custom-built Bitcoin custody platform. Our whitepaper describes how we're different: custodiabank.com/f/custody-w…
Amr retweeted
Chainalysis is claiming that the person who typed all this is not a Bitcoin expert.
diyhpl.us/wiki/transcripts/
Do you think Bryan Bishop is a "Bitcoin expert" as regards the Chainalysis case?
52%Yes
10%No
38%I've never met Bryan
386 votes • Final results Amr retweeted
1/ @CustodiaBank is live!
Last month, we began to serve external customers. For helping us reach this milestone, special thanks go to our team, our investors and all our partners--especially in the face of regulatory pressure on the #digitalasset industry.
Amr retweeted
Although this post was written in response to a specific event, it contains a detailed and technical discussion on error handling in MPC. I don't think that such a discussion appears elsewhere, so I hope that it will be helpful in general.
Replying to @CoinbaseDev
6/ For a more detailed understanding of this issue and our response, we invite you to read our blog post. It provides a comprehensive overview of our ongoing efforts to ensure the highest level of security for our users.
Read the full story here: coinbase.com/blog/the-subtle…
Amr retweeted
Dropping #Downfall, exploiting speculative forwarding of 'Gather' instruction to steal data from hardware registers. #MeltdownSequel
- Practical to exploit (POC/Demo)
- Defeat all isolation boundaries (OS, VM, SGX)
- Bypass all Meltdown/MDS mitigations.
downfall.page
Amr retweeted
We have just uploaded to ePrint a major update to our multiparty ECDSA paper from CCS18. The main changes are in the proof, and we have moved away from Paillier to OT based multiplication. (And new co-authors.) eprint.iacr.org/2018/987.pdf



























