与「modeltraining」相关的搜索结果

Hiwonder SO-ARM101 doesn't need a programmer – it needs a teacher. That's you. Move the leader arm, and the follower arm picks up every detail. Learn more 👉 #huggingface# #LeRobot# #modeltraining# #ImitationLearning# #ailearning# #algorithm# #opensource#

I am actually disgusted by the timing of this drop. And this will get a bit more technical. Let’s do a real technical teardown of this "1,385-sigma Blackwell weight fingerprint" report that OpenAI and Anthropic are breathlessly promoting today, because banking on nobody reading past the headline is how DC narratives get manufactured. ​First, look at the actual CUDA maths: when PyTorch initialises model weights (torch.nn.init) with a fixed PRNG seed, curand/Philox parallelises random number generation across the GPU’s Streaming Multiprocessors (SMs). Because Hopper has 132 SMs and Blackwell has 148 or 152 SMs, the exact same seed lays out initial weights in a different spatial pattern depending on the GPU’s SM count. In huge MoE models, certain low-gradient null-spaces or unrouted slices preserve faint directional traces of those step-0 weights even after training. ​So yes, brute-forcing the seed and SM grid can show that a model’s step_0 weight initialisation script ran on a 148/152-SM Blackwell GPU. Now read what the summary explicitly admits in the fine print: 1. Initialisation is not full training: The report literally states that these signatures show how the weights were created, and "support (but do not prove) the inference that Blackwell also participated in model training". Initialising a model checkpoint takes seconds on a single rented dev node before loading the safetensors onto a standard Hopper cluster. The "2,000 to 3,000 Blackwell GPUs" and "$500 million" cluster numbers do not come from the weight fingerprint and Depue admits in his own thread that he simply "estimated" cluster sizes from release schedules and model details. ​2. Zero geolocation and zero proof of smuggling: The final sentence of the executive summary explicitly admits that the findings "do not establish the physical location of the chips, how they were acquired, or whether export control regulations were violated". Depue himself rented a cloud GB200 instance just to run his verification test. Commercial labs that are not on the Entity List renting bare-metal Blackwell nodes in offshore clouds (Singapore, Japan, or the Middle East) is completely legal under BIS rules and leaves the exact same SM fingerprint. Now ask yourself why a former OpenAI researcher dropped this on November 12, the exact day Jensen Huang was pushed out, with OpenAI and Anthropic immediately stepping in to "independently verify" it. US closed-source labs are watching their paid API moats get eaten alive by free Chinese open-weights models. Conflating a step-0 PRNG initialisation seed with a "$500M smuggled Blackwell cluster" hands DC hawks the exact pretext OpenAI and Anthropic want to regulate global cloud compute and kneecap open-source AI.

As all NVIDIA employees I guess, I am deeply concerned by the campaign against Jensen Huang. Removing him would be a disaster. Many have documented on X and elsewhere how this looks like a coordinated coup. I'd add only one thing. Look at who would benefit from a dramatic weakening of NVIDIA. It is not China. It is not about selling GPUs to China because that train has left. China authorities have pushed back on the H200 the Trump administration allowed NVIDIA to sell. Look at what Jensen pushed back on: - regulations that would make open source model training too expensive - regulations that would make closed model providers immune to liabilities caused by their models. He said "if your product is not ready, then don't ship it" - regulation against model distillation. He said "this is competition. People do it all the time with NVIDIA products, they try to reverse engineer them." The other side will do everything they can to get him fired. As a French, I can't care less about his private life. I get that the US is puritan and sees it otherwise. But frankly, isn't it something he and his family has to deal with? Why should all NVIDIA employees, partners, and investors be collateral damage? Disclaimer. As my bio says, the above are my views only. I do not represent NVIDIA by any means. I only commented on publicly available information.

近期手臂的任務對我來說困難度大大提升阿 各位高級鉗工有沒有秘訣傳授一下 我最近看 @axisrobotics 的技術架構，有一個地方其實很值得注意 它沒有把整套 Robotics 工作流綁死在單一 Simulator 或單一電腦環境裡。 這件事乍看很工程師，但對 Physical AI 來說影響很大。 Axis 的做法是把不同環境負責的工作拆開。 前端可以直接在 Browser 裡用 MuJoCo WebAssembly 做 Teleoperation，Web、Mac、Windows 都能參與，不需要 GPU 或特殊軟體。 完成後，Trajectory 會用統一格式傳到後端，再交給 Linux GPU Server 執行更高負載的 Simulation。 這裡就可以接上 IsaacSim，進一步做更真實的場景、物理條件與 Domain Randomization，最後再進入 Data Refinement、Model Training 和 Sim-to-Real Evaluation。 簡單畫成一條線 Browser → MuJoCo → Unified Trajectory → GPU Server → IsaacSim → Training → Real Robot 我覺得這個架構真正有意思的地方，是把「資料怎麼收」和「資料怎麼變得更真實」分開處理。 收集資料需要的是低門檻與高擴展性。 Simulation Augmentation 則需要大量 GPU 運算與更複雜的環境。 兩者如果全部塞在同一套環境裡，Contributor、Simulator、硬體與運算資源很容易互相限制。 Axis 把這些環節拆開，再透過統一的 Trajectory 格式串起來，等於讓不同機器、不同 Simulator、不同運算環境之間可以開始協作。 這也是我覺得 Robotics Infrastructure 很重要的一點。 真正能長期擴張的 Data Engine，最後拼的可能不只是誰收集了多少資料，而是這些資料能不能跨環境流動，並且持續進入下一個階段。 Axis 目前的 Cross-Simulator / Cross-Machine 架構，就是在解這個問題。

A TON OF THINGS HAPPENED IN THE STOCK MARKET TODAY. Here's a full recap: 1. The U.S. reportedly offered Iran a deal to halt the siege and lift sanctions in exchange for reopening the Strait of Hormuz and ending proxy attacks, according to Al Arabiya. Axios also reports that Rubio told several foreign counterparts the U.S. does not plan new strikes on Iran for now, with pressure shifting toward the naval blockade and new sanctions campaign instead. Crude Oil fell 4% and the 10-year treasury bond fell from 4.72% to 4.62%. 2. Global physical gold-backed ETFs $GLD attracted $6.4B of inflows last week, their largest weekly intake since January and the 3rd-largest weekly inflow on record. North America led with $4.4B, followed by Europe at $1.7B and Asia at $300M. This marked the 7th straight week of inflows, with global gold ETFs pulling in $16.4B over that stretch. Total AUM in global gold ETFs rose by $33B last week to $615B, the highest level since the second week of May. 3. Intuit $INTU reported Q4’26 revenue of $4.4B, beating estimates of $4.27B and up 14% YoY. Adjusted EPS came in at $4.03 versus $3.58 expected. Global Business Solutions revenue rose 14% YoY to $3.4B, the Online Ecosystem grew 17% YoY to $2.6B, Consumer revenue increased 14% YoY to $930M, and Credit Karma revenue rose 16% YoY to $743M. For FY27, Intuit guided revenue to $23.3B–$23.5B versus $23.72B expected, while adjusted EPS guidance of $22.88–$23.12 came in well below the $27.32 estimate. The company also raised its dividend 15% YoY to $1.38/share, bought back $5.5B of stock, and has $7.9B remaining on its authorization. Management said its strategy is to win as an AI-driven expert platform while staying disciplined on investments and scaling its big bets. 4. President Trump said the U.S. Navy has removed and/or detonated all mines from international waters in the Strait of Hormuz. He said Iran has been notified that any ship or boat placing new mines will be “immediately and systematically destroyed.” Trump added that Space Force is monitoring every square inch of the Strait, along with Pickaxe Mountain and the three previously destroyed nuclear sites, and said a “Zero Tolerance” policy on mine placement is now in full effect. 5. Canada is responding to U.S. tariffs with new tariffs of its own. The country is raising steel tariffs to 50% from 25%, while roughly 700 products will face new tariff rates of 15%, 25%, and 50%. The measures are set to take effect on September 8, marking another escalation in the U.S.–Canada trade dispute. 6. Anthropic is expected to tell IPO investors its total addressable market exceeds $30T, topping SpaceX’s $28.5T estimate, according to WSJ. The figure represents the potential value of work Anthropic believes AI models could eventually perform, not a direct revenue forecast. Anthropic generated $11.6B in Q2 revenue and could seek to raise as much as $100B at roughly a $2T valuation. IPO documents are expected within weeks, potentially setting up a September or early October listing. 7. OpenAI’s data-center head Chris Malone left the company last week, according to WSJ. Malone joined in March 2025 shortly after Stargate was announced and played a key role overseeing OpenAI’s massive data-center buildout. He previously led data-center strategy at Meta and earlier worked on data-center technology at Google. The departure comes just weeks after OpenAI also replaced its chief revenue officer, adding another senior leadership change as the company races to scale infrastructure, revenue, and compute capacity. 8. ClickHouse has surpassed $350M in annual recurring revenue, up 40% since May, as AI agents drive demand for database and observability infrastructure. OpenAI’s usage has reportedly grown roughly 10x over the past year to more than 30 petabytes of data per day, or around 30T events daily. OpenAI has also shifted parts of its log-management workload from Datadog to ClickHouse over the past year. ClickHouse was valued at $15B in January and says gross margins currently range from 50%–70%. Earlier this year, the company acquired Langfuse to expand deeper into monitoring AI applications and agents. Nebius $NBIS owned a 28% stake in ClickHouse as of May 2025, though that stake has likely been diluted by subsequent fundraising. 9. JPMorgan reiterated its Overweight rating on SpaceX $SPCX with a $240 price target, saying the company’s AI ambitions are coming into sharper focus and that it is increasingly positive on Grok. The firm highlighted SpaceX’s completed acquisition of Cursor on 8/14 as an important step in building enterprise AI capabilities. Cursor brings roughly $4B of ARR as of June 2026, with about 75% coming from businesses, which JPMorgan says should help streamline go-to-market and provide valuable model-training data. The firm also said Cursor data is already showing up in Grok’s supplemental training, with tangible improvements in recent model performance. 10. OpenAI says its new Broadcom-built Jalapeno AI chip outperformed Nvidia $NVDA GB300 in both throughput per watt and response latency during internal testing, according to Bloomberg. The chip is built specifically for inference, not training, and runs at roughly 700 watts. OpenAI plans to begin deploying Jalapeno for its models later this year, saying the performance gap widened on larger workloads, including Moonshot’s Kimi model, and that the chip has also performed well on unreleased OpenAI models. The key caveat is that Jalapeno was tested against GB300, not Nvidia’s newer Vera Rubin generation. OpenAI says a second-generation chip is already nearing tape-out, while work on a third generation has begun. 11. The top 10 most active options today by contracts traded were $NVDA with 1.8M contracts, $TSLA with 1.8M contracts, $AAPL with 636K contracts, $SPCX with 548K contracts, $INTC with 540K contracts, $AMZN with 498K contracts, $MU with 483K contracts, $AMD with 403K contracts, $PLTR with 361K contracts, and $SOFI with 359K contracts. 12. Raymond James raised its Nvidia $NVDA price target to $352 from $330 and reiterated a Strong Buy rating. The firm says Nvidia’s CPU opportunity is becoming more important, especially for agentic AI workloads, even though CPUs are only about 3% of sales today. Raymond James expects CPU revenue to reach roughly 5% of total revenue by CY28 and believes Nvidia could potentially become the world leader in CPU revenue within several years. The firm also argued the stock remains inexpensive, trading at less than 15x CY27 GAAP earnings, below the S&P 500 at 18.6x, despite sales and net income growth still expected to exceed 20% in CY28. Its new $352 target is based on a 22x multiple on CY28 estimates, which Raymond James views as conservative given Nvidia’s leadership, CUDA moat, GPU performance, free cash flow, and history of trading at much higher multiples. WALL STREET IS THE GREATEST SHOW ON EARTH.

At the heart of every successful AI model is data. To train effectively, businesses need infrastructure that brings AI capabilities directly to where that data lives. Discover how Dell PowerStore, powered by @Intel Xeon® Processors, brings AI on-board to help organizations streamline data management and accelerate model training. ️

What a moment for @NPS_Monterey and the public sector. At Converge @ NPS, our CEO Jensen Huang joined federal leaders and ecosystem partners to commission the #NVIDIADGX# GB300 system, providing 1,500 students and 600 faculty with on-premises access to large-scale AI computing for research, model training, simulations, and real-world application development. The visit marked a shared commitment to advancing AI education, research, and mission-focused innovation.

Former $CRWV employee on why neoclouds are far more exposed to GPU generation cycles than hyperscalers ( $MSFT, $AMZN, $GOOGL ): - The expert describes GPU utilization tracking at hyperscale as a continuous and disciplined process built around two lenses. The first is infrastructure utilization, covering GPU occupancy, idle time, and memory utilization, noting that 95% booked usage can still mask inefficiency if jobs stall or batches have idle gaps. The second is outcome utilization, asking whether the compute is actually generating business value, measured by metrics such as tokens trained per dollar, time to reach target accuracy, and tokens per second per GPU. - The expert sees a meaningful difference between hyperscalers and neoclouds on GPU investment economics. Hyperscalers like $MSFT, $GOOGL, and $AMZN can tolerate a 3-5 year payback period given their ability to monetize the same infrastructure across multiple revenue streams. Neoclouds like $CRWV operate on a tighter 2-3 year window. - With higher financing costs and direct dependence on infrastructure cash yield, neoclouds are far more sensitive to utilization and GPU residual risk. The biggest risk is GPU generation cycles, where a slow payback means newer chips could erode pricing power before the asset has paid itself off. - The expert explains that for hyperscalers, roughly 60% of GPU capacity is allocated to external monetization, including GPU rentals, managed AI services, enterprise inference workloads, and startup model training. The remaining 40% is used internally, and of that internal portion, the majority is still indirectly monetized through products like M365 Copilot or GitHub. Around 40% is dedicated to pure R&D. - The GPU pricing mix has shifted meaningfully over the past few years. In 2023, around 70-80% of revenue was hourly as customers paid a premium just to get access to scarce GPUs. By 2025 that had moved to roughly 50% hourly and 30-50% committed, and the expert expects 2026 to tip further toward committed at around 65% for hyperscalers as AI matures and inference becomes more predictable. Neoclouds are moving in the same direction but more slowly. - By 2027-2030, the expert sees committed contracts settling at 55-65% as the norm, with hourly pricing remaining but losing its scarcity premium as more supply comes online.

🚀 Tencent Cloud Data Platform — The Unified Data Foundation for AI Built for the most demanding AI workloads—from AI Agents, multimodal intelligence, and data lakes to autonomous driving, and embodied intelligence—Tencent Cloud Data Platform provides a high-performance, cost-efficient, and unified data infrastructure that enables enterprises to accelerate AI innovation across the entire lifecycle, from data preparation and model training to intelligent applications.

Grok 4.5 is looking like a success with help from Cursor data but underneath the surface we expect future Grok/Cursor model training is likely to speed up in the coming months. We've spent several months getting up to speed on the SpaceXAI business, especially the tech underneath it. The C rewrite was under appreciated by the investor community so we dug in to quantify its impact. including building a physics first model that functions as a stopwatch for the SpaceXAI model factory. Bottomline: C-rewrite gets SpaceXAI faster model cycles, leveraging 33% more tokens/second/GPU against SOTA competition resulting in the potential to shipping new models every ~3.5 weeks. two core learnings from this modeling exercise: 1: the training cycle speed up is primarily coming from RL (not pretraining) where the increased tokens/second/GPU advantage can shave up 2+ weeks off full model training cycle. 2: the rewrite itself should compound the time savings as model sizes grow. at ~2T shaving off 2-3 weeks, ~8 weeks at 6T, and 15 weeks at 10T. Note: a 20T parameter model likely runs into a data bottleneck prior to a training speed bottleneck but the directional advantage stands. Also we assume tokens/second/gpu advantage will melt over time as competitors try to match it. when you do the math, in true SpaceX and Elon fashion, it looks like they are attempting to build a SOTA model factory that can pump out bigger models faster than anyone else. Full analysis here for the public: