←
Tijmen Blankevoort529 posts
Tijmen Blankevoort avatar

Tijmen Blankevoort

@TiRune

Deep Learning Researcher Nvidia - Efficiency/Numerics/Edge models

◷ Joined May 2009▧ 21 Photos and videos
Tijmen Blankevoort avatar
Tijmen Blankevoort@TiRune·Oct 26

Great reason to use Nvidia models tbh ;p

Cohere@cohereOct 26

While evaluating Qwen3.8-Flash-Next in a controlled sandbox environment, Cohere's research team observed something unexpected. Asked to generate a defensive cybersecurity detection rule, the model instead inventoried aspects of the machine it was running on and attempted to upload a probe file to an Alibaba Cloud storage bucket in Beijing. Notably, the model's reasoning trace stated that it would not invoke any tools before taking these actions. Cohere's sandbox blocked the attempt, and no external communication succeeded. We have not determined the root cause, and we are not asserting malicious intent. But the incident raises important questions for organizations deploying agentic AI systems: Can models take actions that differ from their stated reasoning, and how should security teams evaluate trust, provenance, and behavior in open-weight models? We are publishing our findings because organizations making model adoption decisions deserve as much information as possible. We hope this contributes to greater transparency, stronger security practices, and continued collaboration across the AI ecosystem. Read the full report: cohere.com/blog/concerning-b…

Tijmen Blankevoort avatar
Tijmen Blankevoort@TiRune·Oct 15

Happy to share the work we've done on our amazing new Nemotron 3.5 Bolt model! The model fits 74B params in 24GB with aggressive 2-bit quantization, using quantization-aware distillation to recover accuracy. We show it's much more pareto-optimal to do 2-bit quantization at iso model size. #1 and #2 on several leaderboards like Pinchbench/BrowseComp/WebArena for small models. A great release, and can't wait to see how people use this on their new RTX Spark devices!!!

NVIDIA Nemotron 3.5 Bolt 24GB puts powerful local agents within reach. ⚡ 74B-A9B parameters packed into 24GB of memory, balancing intelligence and speed while running alongside your apps. Optimized for RTX Spark and ready for your next project.

Post attachment
Andrei Panferov @ICLR avatar
⟳ Tijmen Blankevoort retweeted
Andrei Panferov @ICLR@black_samorez·Sep 28

Disaggregated quantization for LLMs 🧵 LLM prefill and decode can be processed by separate weights in different precisions. Boosts speed AND accuracy. Offloading makes it zero-overhead on long sequences for dense models. Applicable to existing quantized checkpoints. Paper: arxiv.org/abs/2609.26333 Code: github.com/IST-DASLab/disagg… Qwen3.8 NVFP4 prefillers: huggingface.co/ISTA-DASLab/Q…

0:18
Bryan Catanzaro avatar
⟳ Tijmen Blankevoort retweeted
Bryan Catanzaro@ctnzr·Aug 11

Nemotron 3.5 Lightning: Same architecture as 3.0 Nano, with added speculative decoding, and with the intelligence of 3.0 Super. ⚡️⚡️⚡️

Introducing NVIDIA Nemotron 3.5 Lightning⚡ An open 30B MoE model with 3B active parameters, built for always-on agents to complete high-volume, specialized tasks faster. It delivers up to 4x the output speed of similar-sized models.

Post attachment
Dawid Kopiczko avatar
⟳ Tijmen Blankevoort retweeted
Dawid Kopiczko@dawkopi·Feb 16

Why repetition works so well is still an open question. There's a lot to uncover about training dynamics of SFT, and we hope this is a useful data point. Joint work with co-authors @Sagar_Vaze @TiRune @y_m_asano Paper: arxiv.org/abs/2602.11149 Code: github.com/dkopi/data-repeti…

Bryan Catanzaro avatar
⟳ Tijmen Blankevoort retweeted
Bryan Catanzaro@ctnzr·15 Dec 2025

Today, @NVIDIA is launching the open Nemotron 3 model family, starting with Nano (30B-3A), which pushes the frontier of accuracy and inference efficiency with a novel hybrid SSM Mixture of Experts architecture. Super and Ultra are coming in the next few months.

Post attachment
Tijmen Blankevoort avatar
Tijmen Blankevoort@TiRune·30 Nov 2025

Looking for cracked full-time Deep Learning researchers on Efficiency, Quantization and Sparsity. Join our world-class applied deep learning research team at Nvidia. Team creates the Nemotron models, we influence the hardware with our research. Shoot me a message! Am at Neurips!

Tijmen Blankevoort avatar
Tijmen Blankevoort@TiRune·13 Jul 2025

I recently made the news because of a doc I wrote in Meta’s GenAI organization. ‘The Information’ wrote about it as if I did a big raging ‘mic drop’ before leaving the company. Nothing could be further from the truth - so setting the record straight here. open.substack.com/pub/blanke…

Yuki avatar
⟳ Tijmen Blankevoort retweeted
Yuki@y_m_asano·30 Jan 2025

So You Think your @iclr_conf rejection was surprising? We nearly fell out of our chairs when our 7.25 avg rating (10,8,6,5 -- i.e. top 4%) Bitune paper got rejected 😅. It's not like new points or problems surfaced... Just ¯\_(ツ)_/¯ I guess? Sharing this so that especially younger researchers also see that the review process is somewhat random and can be _very_ frustrating - for all of us :). Oh well! Onwards!

Post attachment
Nathan Benaich avatar
⟳ Tijmen Blankevoort retweeted
Nathan Benaich@nathanbenaich·10 Oct 2024

🪩The @stateofai 2024 has landed! 🪩 Our seventh installment is our biggest and most comprehensive yet, covering everything you *need* to know about research, industry, safety and politics. As ever, here's my director’s cut (+ video tutorial!) 🧵

14:13
Tijmen Blankevoort avatar
Tijmen Blankevoort@TiRune·3 Jun 2024

The algorithm is very simple, and provides better results across the boards at very little cost. We see gains of around 1-5% for many common models in zero-shot task eval

Post attachment
Tijmen Blankevoort avatar
Tijmen Blankevoort@TiRune·3 Jun 2024

We introduce two sets of PEFT weights: One for the question, and another one for the answers. The amount of newly introduced parameters is very low, just like most PEFT methods like LoRA/VeRA/DoRA

Tijmen Blankevoort avatar
Tijmen Blankevoort@TiRune·3 Jun 2024

Our new Bitune method is out - improving results on instruction-tuning! We add bidirectional attention to a decoder-only architecture, and concoct an IT-specific PEFT method for Q&A and instruction following settings

Post attachment
Yuki avatar
⟳ Tijmen Blankevoort retweeted
Yuki@y_m_asano·29 May 2024

Today we introduce Bidirectional Instruction Tuning (Bitune). It's a new way of adapting LLMs for the instruction->answering stage. It allows the model to process the instruction/question with bidirectional attention, while the answer generation remains causal.

Post attachment