Jiaming Tang (@jmtang42) profile picture

Ph.D. student @MIT. Prev: Intern @physical_int, @NVIDIA. I am interested in MLSys & Algo.

Cambridge, MA
Joined January 2023
Everyone's been showing Fable and GPT-6 Astra driving robot arms. We trained an open 27B Qwen model to do it, purely from its own attempts in simulation, with zero demonstrations. And it can run locally at 220+ tok/s on an RTX 5090. On a real robot it had never touched, it now succeeds in 18 of 30 tests, up from 0 before training. It also copies a painting and designs its own chemistry experiment. It generalizes and follows instructions better than π0.5, and on LIBERO it comes close to Astra max. All without losing its general capabilities, like math and coding. Built on Qwen3.8-27B. We've open-sourced the model weights, code, and full training infra! Check out our blog at mit-han-lab.github.io/pear-b….
52
168
50
1,224
110,581
Jiaming Tang retweeted
🔥 Serving coding agents well needs a high-performance prefill engine, but existing systems fall short. 🚀 Introducing SSEngine-GLM5.3-Flash: an open-source prefill engine for GLM-5.3-Flash on four B200 GPUs. It combines pipeline parallelism, a tiered prefix cache across GPU memory, DRAM and SSD, and a shortest-remaining-prefill-first scheduler. 📈 On real hardware, it sustains 3.1× (AgentX) and 3.2× (TraceLab) the request rate of any vLLM or SGLang setup we tested. 🧵 More in the thread below (1/8)
1
4
1
5
366
Jiaming Tang retweeted
What does RoboDojo really simulate? We followed it from the real world to its score, one link at a time. At almost every step, it simulates something other than what it says. Same bottle, same grasp. As shipped, it tips 78° in the hand. After our fix, it stays level.
6
16
13
136
28,215
Jiaming Tang 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
213
15,834
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 👇
18
50
11
216
46,118
Jiaming Tang retweeted
🤖 What excites me about this work is the simplicity of the paradigm. A reasoning model uses a robot much as we might imagine: it observes through cameras, reasons about what to do, and takes action. Through RL entirely in simulation, the model learns to put its existing knowledge and reasoning to work on real robots—with no demonstrations. These skills transfer across tasks and robots while preserving the model’s broader capabilities. Could Robot-Use RL (RURL) become part of foundation-model post-training, with simulators providing experience at scale? We’re excited to see how far this can go! Many thanks to our collaborators for making this work possible! Try it out and share your feedback: mit-han-lab.github.io/pear-b…
Everyone's been showing Fable and GPT-6 Astra driving robot arms. We trained an open 27B Qwen model to do it, purely from its own attempts in simulation, with zero demonstrations. And it can run locally at 220+ tok/s on an RTX 5090. On a real robot it had never touched, it now succeeds in 18 of 30 tests, up from 0 before training. It also copies a painting and designs its own chemistry experiment. It generalizes and follows instructions better than π0.5, and on LIBERO it comes close to Astra max. All without losing its general capabilities, like math and coding. Built on Qwen3.8-27B. We've open-sourced the model weights, code, and full training infra! Check out our blog at mit-han-lab.github.io/pear-b….
5
5
34
1,952
You’ve seen Astra draw. This time, it’s an open-source 27B reasoning model we trained! RL entirely in simulation. No demonstrations. No agent harness. Native robot use. I still can‘t believe what we achieved, from an open-source foundation model with almost no native robot use capability to perform well on the real robot. This is really an exciting moment for me as it means a fundamentally different path from existing robot foundation models, directly addressing their biggest limitations. Robot foundation models have largely relied on learning manipulation from robot data, which is painfully scarce. Robot-use agent inherits the LLM’s native world knowledge and reasoning capability, and we just unleash and put them to work on robot. With robot demonstrations so scarce, being able to draw on that existing intelligence matters enormously. The exceptional instruction following and generalization have been striking to watch. Capabilities we’ve come to expect from LLMs suddenly feel remarkable when they show up in a physical robot. In our experiments, more compute and more simulation environments keep translating into stronger capabilities. We also see transfer across tasks, across robots, and from simulation to the real world. That makes the direction feel tangible: scalable simulation compute can translate into transferable robot capabilities. There’s something deeply encouraging about seeing a path where investing more resources keeps moving the needle. We’re sharing our observations and results, and open-sourcing the model, training infrastructure, and deployment stack so everyone can build on them. We’ve also laid out the limitations in detail. Each one points to an open research question, and together they capture much of what we believe are promising directions in the future. Excited to see what others discover—and build. Enjoy! mit-han-lab.github.io/pear-b…
Everyone's been showing Fable and GPT-6 Astra driving robot arms. We trained an open 27B Qwen model to do it, purely from its own attempts in simulation, with zero demonstrations. And it can run locally at 220+ tok/s on an RTX 5090. On a real robot it had never touched, it now succeeds in 18 of 30 tests, up from 0 before training. It also copies a painting and designs its own chemistry experiment. It generalizes and follows instructions better than π0.5, and on LIBERO it comes close to Astra max. All without losing its general capabilities, like math and coding. Built on Qwen3.8-27B. We've open-sourced the model weights, code, and full training infra! Check out our blog at mit-han-lab.github.io/pear-b….
6
7
1
83
5,778
RL on robot tasks leaves its other skills in place: no measurable loss across nine reasoning, coding and vision benchmarks. It still scores 98.5 on AIME 2026, 89.7 on GPQA Diamond and 88.4 on LiveCodeBench.
1
15
2,377
Jiaming Tang retweeted
Nunchux now leads 10 @ArtificialAnlys leaderboards for image and video generation speed, covering MiniMax-H3, LTX-2.3, Qwen-Image, and FLUX series. 🚀 These results come from building the Nunchux Model Optimizer and Nunchux Inference Engine together. We want to make image and video generation interactive and affordable, so you can try an idea and refine it as you go. The waitlist is gone. Nunchux Modelverse is open to everyone, with 80+ models. More models and categories are on the way. Sign up today for free credits and generate faster for less. Full results and comparisons on our blog: nunchux.ai/blog/nunchux-top-… Special thanks to @ArtificialAnlys team for onboarding us to their benchmarks. #Nunchux #AIInference #MultimodalAI #ImageGeneration #VideoGeneration
24
23
8
88
152,596
Jiaming Tang retweeted
A takeaway from my #Runtime talk recently: drafting one token at a time becomes its own bottleneck. With #DFlash, we draft a whole block in parallel. That makes room for a deeper, more accurate drafter while keeping latency low. The target model still verifies every token. At @inco_ai, we’re building on this work to make inference faster for AI applications and agents.
8
6
2
40
3,312
Jiaming Tang retweeted
Inco is going to #SGLangSummit 2026 🧡 Our open-model endpoints run several times faster than most providers on @ArtificialAnlys. Our co-founder @ZhijianLiu will lead a breakout session on #DFlash and what it takes to build inference for the agentic era, at scale. 🎙️ DFlash and Beyond: How Inco Runs the Fastest Open-Model Endpoints 📍 Nov 12–13, SF 🎟️ luma.com/sglang-summit Coming to the summit? Stop by, say hi, and let's talk about your inference stack.
3
6
2
46
6,485
Jiaming Tang retweeted
MiniMax M3.1 Flash is out, and we've open-sourced a DFlash 2 drafter for it under Apache 2.0. It makes generation up to 3.5× faster, and it's lossless. Congrats to @MiniMax_AI on the release! huggingface.co/incoai/MiniMa…
MiniMax M3.1 Flash weights are now public. 428B parameters total, 23B activated per token. Agentic performance: A generational leap in agentic tasks, coding, and cybersecurity — 76.2 on BankerToolBench, 73.7% on DeepSWE v1.1, 61.7 on OSWorld V2, 29.6 on Agents' Last Exam, and 82.2 on CyberGym. Efficient by design: MiniMax Sparse Attention and DSpark speculative decoding improve inference efficiency. Hugging Face: huggingface.co/MiniMaxAI/Min… ModelScope: modelscope.cn/models/MiniMax… API: platform.minimax.io M Plan: minimax.io/m-planMiniMax MCode for desktop: agent.minimax.io/download
8
16
11
138
23,537
Jiaming Tang retweeted
Splash 1.3.0 is out! At launch we focused on decode speed. This time we worked on the wait before the first token. Watch the part where we come back to a project. With SSD offloading on, the first token takes about a second instead of 19.
Splash 1.3.0 is out, and it's about the wait before your local agent starts answering ⚡ If you switch projects and come back, SSD offloading brings the first token down from 19 seconds to 1. New tasks start 1.5× faster, thanks to the Neural Engine. All on a 24 GB M6 Mac.
3
10
1
51
68,374
Jiaming Tang retweeted
Splash 1.3.0 is out, and it's about the wait before your local agent starts answering ⚡ If you switch projects and come back, SSD offloading brings the first token down from 19 seconds to 1. New tasks start 1.5× faster, thanks to the Neural Engine. All on a 24 GB M6 Mac.
6
26
5
125
96,035
Jiaming Tang retweeted
31× more video per GPU, at 1/15 the cost ⚡ That's @linewiseAI's video agent on Inco vs. their own pipeline. The agent is thorough. Before it calls a moment idle, it checks what came before and after, reads the sensors, and zooms in. Lots of model calls, and we make them fast.
One week of our video data: 16,190 hours. A full-time worker would need 8 years to watch it. Our agent-native infra analyzes it in a day, with inference on @inco_ai: 31× the throughput per GPU at 1/15 the cost per hour of video.
1
5
2
35
81,509