ทวีตที่ปักหมุด
罗杰斯
544 posts
罗杰斯
@dhbrojas
AGI @ 池畔 🏖️ Prev. https://t.co/vrJX6VP8I0, 清华大学
Paris เข้าร่วม Nisan 2020
1.1K กำลังติดตาม435 ผู้ติดตาม
罗杰斯 รีทวีตแล้ว
@jayair I'm an emo kid, non conforming as can be. You'd be non conforming too if you looked just like me.
English
罗杰斯 รีทวีตแล้ว
ML perf is in its JS framework era, i love it.
Bohan Hou@bohanhou1998
We release TIRx today, a minimal compiler stack and hardware-native DSL for frontier ML kernels, built around storage-first tensor layouts and reusable tile primitives. tvm.apache.org/2026/06/22/tirx On NVIDIA B200, TIRx delivers up to ~1.08× over cuBLASLt on dense GEMM, outperforms DeepGEMM on all FP8 blockwise workloads with up to ~1.09× speedup, keeps FlashAttention-4 (FA4) typically within ~±2% of CuTeDSL, and remains competitive with cuBLASLt/FlashInfer on NVFP4 GEMM. Through our past experiences building frontier ML kernels, megakernels, and agentic kernel systems, we kept seeing the same boundary problem: new operators and new hardware require new optimization strategies that often break old programming models or compiler passes. TIRx builds on top of Apache TVM and moves toward a simple goal: let users and agents express the best-performing program, even for future hardware generations, while keeping the engineering effort for new kernels and new hardware as low as possible.
English
I havent tried buying a $20k personal inference rig yet because I plan to buy a $5k one that outperforms it in a couple years
罗杰斯@dhbrojas
Trust me bro, it's only $20K bro, you only need four of them bro, you can run GLM 5.2 REAP INT2 at 20 tokens/s bro
English
New Z.ai office logo next to Tsinghua University
Z.ai@Zai_org
Introducing GLM-5.2: Frontier Intelligence, Open Weights - Significant improvements in coding and agentic tasks - Strong long-horizon capabilities with a 1M context window - Two levels of reasoning effort: GLM-5.2 (max) pushes the limits, while GLM-5.2 (high) strikes a strong balance between performance and token efficiency - MIT-licensed open weights - Same API pricing as GLM-5.1 Tech Blog: z.ai/blog/glm-5.2 Weights: huggingface.co/zai-org/GLM-5.2 API: docs.z.ai/guides/llm/glm… Coding Plan: z.ai/subscribe Chat: chat.z.ai
English
guy who buys $20k of hardware to avoid paying these astronomical prices
Scenic Oaks, TX 🇺🇸 English
罗杰斯 รีทวีตแล้ว
GLM 5.2 is one *of the* greatest gap reductions ever, but I think it is *the* greatest show of benchmark solidity from an open model claiming SoTA ever. Normally, you have some variety of the bad old Qwen pattern: headline benchmarks are SoTA+, new OOD ones are ≈8 months behind, and real experience is spiky, competitive in places, but usually ≈1 year behind, and sometimes utterly falling apart. Knock on it and hear the hollow sound. Yes, even DeepSeek.
Not so here. There's no progressive decay. It's "Opus 4.5-4.7ish" throughout, in anything of value that you throw at it. It is the first truly, completely solid Chinese model. A phase change, I hope.
Elliot Arledge@elliotarledge
Beyond the megakernel, a 6-problem hard CUDA/Triton deck. Speedup is over torch.compile (a strong baseline, not naive PyTorch). Paged attention is where compile falls down and a real kernel runs away with it: Opus 4.8 hits 56.8x on B200.
English
People always dunk on France and our issues with #clim but not many know about Fraîcheur de Paris.
Paris has a mutualized cooling system that exchanges heat with the cool Seine and pumps through the entire city.
Individual AC is just so stupid compared to that: ~35% less electricity, ~50% less CO₂, ~90% less refrigerant emissions and much lower cost once installed.
English
罗杰斯 รีทวีตแล้ว
