Kunjun Li (@li_kunjun) header
Kunjun Li (@li_kunjun) profile picture
Berkeley, CA
Joined November 2024
Photo by Kunjun Li (@li_kunjun)
Reasoning models can now drive robots. But the robot sits still while the model thinks. SpecPilot now speeds up agentic robotics. On PEAR (Qwen3.8-27B as a robot policy), the same LIBERO task finishes in 49 s instead of 189 s. 3.84× faster task completion 71 → 241 tok/s decode No training 🤖
RSI starts with infra 🔁 mlsys is in a golden age. agents are great researchers now. with tokens, "low-hanging fruit" is everywhere. agents optimize speculative decoding for parallel drafters (DFlash2, DSpark) without training or quality loss. their 12-bit format beats cuBLAS.
13
21
4
214
16,375
Kunjun Li retweeted
Dear researchers: Did you know that your H100 can actually ray-trace like a gaming card? 🎮 for a long time, people have assumed that compute cards like the H100 can only compute, and that you need a graphics card like the RTX 5090 to render. we show that's wrong. 🚨Meet RTX H100. Blender Cycles on the Open Data benchmark (Blender 5.3.0): ours 48,641. the RTX 5090 is at 19,202, and a stock H100 gets only 7,041. our original goal was to match the 5090. we ended up at 2.5x. our neural rendering approach is lossy. we test within a reasonable loss, and every result passes an image check. we think nobody doing this before was an accident: graphics people rarely touch compute cards, and people who have compute cards rarely know graphics. robotic simulators need fast rendering. now a general-purpose compute card can do it, so you don't need heterogeneous compute or a complicated cluster. we'll extend this to more hardware next. almost ready for GTA 6 on the cluster 🤣 Best, Wenhao
28
34
14
218
24,212
Kunjun Li retweeted
12-bit lossless also found by Opus 5.5 in our recent SpecPilot project. Seems low-hanging fruit 🍉;) paper: arxiv.org/pdf/2611.02501 post: teapawt.org/wenhaocha1/status/2117…
Replying to @xingyudang
The agents found interesting tricks. On Qwen3-8B: a lossless 12-bit weight format (each row spans only a few exponents), decoded in registers on the way to the tensor cores: −15% per token. 19 runs, 442 agent-hours, 675 submissions, with Claude Opus 5.5 and Fable 5.2.
2
4
38
3,178
Kunjun Li retweeted
Exciting to see what DFlash2 and SpecPilot can achieve together. Parallel drafting and systems optimization go hand in hand, with so much more potential to unlock. We believe parallel drafters are the future.
Reasoning models can now drive robots. But the robot sits still while the model thinks. SpecPilot now speeds up agentic robotics. On PEAR (Qwen3.8-27B as a robot policy), the same LIBERO task finishes in 49 s instead of 189 s. 3.84× faster task completion 71 → 241 tok/s decode No training 🤖
1
1
26
1,031
Kunjun Li retweeted
Reasoning models can now drive robots. But the robot sits still while the model thinks. SpecPilot now speeds up agentic robotics. On PEAR (Qwen3.8-27B as a robot policy), the same LIBERO task finishes in 49 s instead of 189 s. 3.84× faster task completion 71 → 241 tok/s decode No training 🤖
RSI starts with infra 🔁 mlsys is in a golden age. agents are great researchers now. with tokens, "low-hanging fruit" is everywhere. agents optimize speculative decoding for parallel drafters (DFlash2, DSpark) without training or quality loss. their 12-bit format beats cuBLAS.
13
21
4
214
16,375
How: PEAR runs on the SpecPilot engine (W12 weights, tree verification, two-level verification), plus an efficient reasoning rule: • accept drafted tokens the target gives p ≥ 0.3 • skip hesitation words ("Wait", "Hmm") • never relax inside a repeated sequence Median speedup over the PEAR baseline: 4.04× across 19 LIBERO episodes.
1
6
865
The reasoning rule alone makes the robot clock 1.87× faster, with the same success rate (81.7% vs 81.7%) over 120 paired episodes. The rule is lossy and validated on LIBERO only. Code and reproduction scripts are in the repo. 🌐 stargazerx0.github.io/specpi… 💻 github.com/StargazerX0/SpecP…
5
720
How: PEAR runs on the SpecPilot engine (W12 weights, tree verification, two-level verification), plus an efficient reasoning rule: • accept drafted tokens the target gives p ≥ 0.3 • skip hesitation words ("Wait", "Hmm") • never relax inside a repeated sequence Median speedup over the PEAR baseline: 4.04× across 19 LIBERO episodes.
1
18
The reasoning rule alone makes the robot clock 1.87× faster, with the same success rate (81.7% vs 81.7%) over 120 paired episodes. The rule is lossy and validated on LIBERO only. Code and reproduction scripts are in the repo. 🌐 stargazerx0.github.io/specpi… 💻 github.com/StargazerX0/SpecP…
14
We breathed life into LLM system. Introducing Evolving Infra⚙️: an inference framework that adapts diverse and dynamic agent workload purely online and retires fixed PD disaggregation. vs. NVIDIA Dynamo on up to 256 GB300 GPUs: ➡️ Up to 4.1× lower latency ➡️ Up to 14× lower P95 TTFT ➡️Up to 2.36× more revenue on real agent traces(Computer Use, Coding and math agents) and opensource models. 🌐 Website: inference-rsi.com 📄 Tech report: alphaxiv.org/en/abs/2611.123… 💻 Code: github.com/mit-han-lab/Evolv… Which agent(e.g. Pi/Openhands) do you want us to accelerate next? Tell us below 👇
21
60
15
271
56,031
Kunjun Li retweeted
no I’m totally wrong. It is ICLR paper acceptance prediction task ;) anyone disagree?
Im sorry but I’m afraid it’s heads or tails bro 😎
2
3
16
2,628
Kunjun Li retweeted
RSI starts with infra 🔁 mlsys is in a golden age. agents are great researchers now. with tokens, "low-hanging fruit" is everywhere. agents optimize speculative decoding for parallel drafters (DFlash2, DSpark) without training or quality loss. their 12-bit format beats cuBLAS.
15
33
10
335
55,755
SpecPilot is out 🚀 we gave coding agents GPUs and one rule: stay lossless. with one request, they more than doubled every released speculative decoding system we tested. DFlash2: 4.46× → 10.87× over AR. no training. plus a lossless 12-bit format whose GEMM beats cuBLAS.
RSI starts with infra 🔁 mlsys is in a golden age. agents are great researchers now. with tokens, "low-hanging fruit" is everywhere. agents optimize speculative decoding for parallel drafters (DFlash2, DSpark) without training or quality loss. their 12-bit format beats cuBLAS.
10
15
2
154
12,838
Kunjun Li retweeted
Full results of Astra, Argon, and Carbon. in this task agents are asked to generate policy weight to play a game at one shot without any interaction.
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math. It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training. We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵 Blog: rikiyat.github.io/blog/weigh…
Readers added context they thought people might want to know
LLM was given weights explicitly as text in prompt and used code/forward passes (per authors' logs) to ablate Qwen looping direction, a known thinking-mode issue per model card. Primarily fixed repetition in that mode, not general 'natively' improving weights/math. rikiyat.github.io/blog/weight-sp… github.com/RikiyaT/lm_und… huggingface.co/Qwen/Qwen3.5-0… huggingface.co/RikiyaT/Qwen3.…
10
14
2
120
21,045