←
San Francisco Tensor Company22 posts
San Francisco Tensor Company avatar

San Francisco Tensor Company

@SfTensor

The San Francisco Tensor Company

◷ Joined September 2025▧ 3 Photos and videos
Ben Koska avatar
⟳ San Francisco Tensor Company retweeted
Ben Koska@BenKoska·6h

we’re hosting gpu poker tomorrow evening at our office: come touch some GPU nodes & IB switches, compliment our rack, have fun and win some B300 hours (no buy in)

Post attachment
Ben Koska avatar
⟳ San Francisco Tensor Company retweeted
Ben Koska@BenKoska·Nov 3

We formalized PTX in Lean so that we can formally prove GPU kernels correct. This way, we can be confident that the kernels our agents wrote are correct, without ever needing to read them. 🧵

Ben Koska avatar
⟳ San Francisco Tensor Company retweeted
Ben Koska@BenKoska·Oct 30

Today, there are 3 classes of compute: On-Demand is reliable but expensive, Reserved requires you to commit to your spend years out and Spot is cheap but can be reclaimed at a moments notice and anything running dies. What if we could just... continue a process on a new host? 🧵

Ben Koska avatar
⟳ San Francisco Tensor Company retweeted
Ben Koska@BenKoska·Sep 26

turns out when you have too many GPUs for your own good you can do stupid stuff such as pre-training a foundation model with constrained decoding on PTX -- that is, it does not speak natural language, it's tokenizer is PTX, it's thoughts are PTX, it's all PTX 🧵

Ben Koska avatar
⟳ San Francisco Tensor Company retweeted
Ben Koska@BenKoska·Sep 25

We're #1 on NVIDIA's SOL-ExecBench across all four tracks. Across 235 real production kernels run on B200s we beat out 583 teams, including teams that have raised hundreds of millions of dollars to optimize kernels. Along the way we learned a lot about agentic optimization 🧵

Post attachment
Ben Koska avatar
⟳ San Francisco Tensor Company retweeted
Ben Koska@BenKoska·Sep 12

We've been working on training our own models for large-scale, long-running optimization tasks. The model is still training, but we asked an early checkpoint to optimize pre-training on 32 MI355x, surprisingly it manages to train faster than NVIDIA NeMo running on 32 GB200s 🧵

Post attachment
Ben Koska avatar
⟳ San Francisco Tensor Company retweeted
Ben Koska@BenKoska·Aug 8

for RL and agents you need fast, isolated sandboxes which is why we've used gVisor for CPU and NVIDIA GPU workloads. as we started post-training models for AMD GPU kernels though, we hit a problem: gVisor has no AMD GPU support. or rather, it didn't, until we built it :)

0:40
Ben Koska avatar
⟳ San Francisco Tensor Company retweeted
Ben Koska@BenKoska·Aug 6

We have Day‑0 support for Xiaomi-Robotics-1 at @SFTensor Let us help you specialize it for your use case

Post attachment

Xiaomi-Robotics-1 is now open source!🎉 Xiaomi-Robotics-1 was pre-trained on more than 100,000 hours of UMI data and post-trained with over 10,000 hours of cross-embodiment data. This open-source release covers the complete pipeline from real-robot post-training to model deployment and also includes evaluation code for benchmarks. 🌐Project Page :robotics.xiaomi.com/xiaomi-r… 💻GitHub: github.com/XiaomiRobotics/Xi… 📎Hugging Face: huggingface.co/collections/X… We will continue exploring the broader possibilities of general-purpose robot models. #Robotics #EmbodiedAI #FoundationModels #RobotLearning #XiaomiAI #XiaomiRobotics

Post attachment
Ben Koska avatar
⟳ San Francisco Tensor Company retweeted
Ben Koska@BenKoska·Aug 5

The biggest advantage of owning the backend end-to-end is that instead of relying on ptxas, you can try different ways of scheduling, such as search & RL. While ptxas beats our beta backend without search, once we turn on search we beat it by ~8 points 🧵

Post attachment

we've long been dissatisfied with the fact that the closest we can get to the hardware on NVIDIA is PTX, with ptxas then making all sorts of decisions (instruction selection, barriers, stall counts, etc.) for us without any way to influence it, so we've been building a custom LLVM backend that let's us create cubins directly from LLVM, which let's us do things such as write inline SASS and define custom scheduling. reply or reach out if you wanna give it a twirl

0:20
Ben Koska avatar
⟳ San Francisco Tensor Company retweeted
Ben Koska@BenKoska·Aug 5

we've long been dissatisfied with the fact that the closest we can get to the hardware on NVIDIA is PTX, with ptxas then making all sorts of decisions (instruction selection, barriers, stall counts, etc.) for us without any way to influence it, so we've been building a custom LLVM backend that let's us create cubins directly from LLVM, which let's us do things such as write inline SASS and define custom scheduling. reply or reach out if you wanna give it a twirl

0:20
Ben Koska avatar
⟳ San Francisco Tensor Company retweeted
Ben Koska@BenKoska·Aug 4

The benefits of AI are dangerously concentrated. Today they accrue to a handful of labs and to one industry, software, whose work happens to look like the internet the models were trained on. Changing that is the single biggest untapped lever for economic growth this decade.🧵

Ben Koska avatar
⟳ San Francisco Tensor Company retweeted
Ben Koska@BenKoska·Aug 4

And because we automatically target whichever hardware is cheapest or best for a given workload, vendors have to compete on performance and price instead of coasting on lock-in. A chip that becomes uncompetitive can no longer hide behind switching costs. It has to win on the merits.

Ben Koska avatar
⟳ San Francisco Tensor Company retweeted
Ben Koska@BenKoska·Aug 4

We built a software model of NVIDIA Blackwell's Tensor Core MMA instruction (tcgen05.mma) that matches real B200 hardware bit-for-bit. Not "close enough for numerics." Every output word, identical. Here's how we reverse-engineered it. 🧵

Post attachment
Ben Koska avatar
⟳ San Francisco Tensor Company retweeted
Ben Koska@BenKoska·Aug 4

We've stopped testing our GPU kernels and started proving them correct instead. A test suite checks the inputs you thought of. A proof covers every input that exists and hands you the exact one that breaks the kernel if it's wrong. Here's how. 🧵

Post attachment
Ben Koska avatar
⟳ San Francisco Tensor Company retweeted
Ben Koska@BenKoska·Jul 30

The moment we saw this letter, we signed. An open ecosystem beats a closed moat every time and is how AI actually reaches every sector. Proud to stand with everyone helping build and accelerate open, American AI

Since launching last week, more than 230 companies and organizations from across the tech sector have signed the "Open Weights and American AI Leadership" open letter. We want to thank these partners for standing up and publicly supporting broader access to AI innovation. A special thanks to @nvidia, @a16z, and @PalantirTech for working with @Microsoft on this effort. These signatories understand that America’s AI leadership will not depend on the success of our frontier models alone, but on our ability to build a strong, secure, and open ecosystem that diffuses AI into every sector. We look forward to continuing to work with our partners and with policymakers to build that open ecosystem in a way that benefits American businesses, empowers American workers, and strengthens the American economy.

Post attachment
Ben Koska avatar
⟳ San Francisco Tensor Company retweeted
Ben Koska@BenKoska·Jul 21

We built a benchmark where the AI agent has to build an inference chip. GPT-5.6 took 3.4 days and 1.7B tokens. The agent has to design, verify and physically place-and-route a working inference accelerator from a single MAC unit to a routed die that decodes tokens bit-exactly.

Post attachment