与「C_real」相关的搜索结果

In 2019, OpenAI announced GPT-2 with this post: Today (~5 years later) you can train your own for ~$672, running on one 8XH100 GPU node for 24 hours. Our latest llm.c post gives the walkthrough in some detail: Incredibly, the costs have come down dramatically over the last 5 years due to improvements in compute hardware (H100 GPUs), software (CUDA, cuBLAS, cuDNN, FlashAttention) and data quality (e.g. the FineWeb-Edu dataset). For this exercise, the algorithm was kept fixed and follows the GPT-2/3 papers. Because llm.c is a direct implementation of GPT training in C/CUDA, the requirements are minimal - there is no need for conda environments, Python interpreters, pip installs, etc. You spin up a cloud GPU node (e.g. on Lambda), optionally install NVIDIA cuDNN, NCCL/MPI, download the .bin data shards, compile and run, and you're stepping in minutes. You then wait 24 hours and enjoy samples about English-speaking Unicorns in the Andes. For me, this is a very nice checkpoint to get to because the entire llm.c project started with me thinking about reproducing GPT-2 for an educational video, getting stuck with some PyTorch things, then rage quitting to just write the whole thing from scratch in C/CUDA. That set me on a longer journey than I anticipated, but it was quite fun, I learned more CUDA, I made friends along the way, and llm.c is really nice now. It's ~5,000 lines of code, it compiles and steps very fast so there is very little waiting around, it has constant memory footprint, it trains in mixed precision, distributed across multi-node with NNCL, it is bitwise deterministic, and hovers around ~50% MFU. So it's quite cute. llm.c couldn't have gotten here without a great group of devs who assembled from the internet, and helped get things to this point, especially ademeure, ngc92, @gordic_aleksa, and rosslwheeler. And thank you to @LambdaAPI for the GPU cycles support. There's still a lot of work left to do. I'm still not 100% happy with the current runs - the evals should be better, the training should be more stable especially at larger model sizes for longer runs. There's a lot of interesting new directions too: fp8 (imminent!), inference, finetuning, multimodal (VQVAE etc.), more modern architectures (Llama/Gemma). The goal of llm.c remains to have a simple, minimal, clean training stack for a full-featured LLM agent, in direct C/CUDA, and companion educational materials to bring many people up to speed in this awesome field. Eye candy: my much longer 400B token GPT-2 run (up from 33B tokens), which went great until 330B (reaching 61% HellaSwag, way above GPT-2 and GPT-3 of this size) and then exploded shortly after this plot, which I am looking into now :)

Since my name came up, a short note. Yes, I played the Bike game in 2023, for about an hour, and had a small piece of one player. It was over two years ago. Hank and Alan have already described what happened that night, and my recollection is roughly the same. I wasn't there for the settlement afterwards, so I won't speak to it. As for C Ye and Britney, that's between them. I wasn't involved and I'm not picking a side. If I can offer one thought: this all feels bigger from inside the poker bubble than it really is. Most people have never heard of it and never will. Settle it privately if you can, and move on. That's the whole statement. I won't be commenting on this again or answering questions about it, so please don't read my silence as agreement or disagreement with anyone. Poker isn't really where my head is these days. Life is full of more interesting things.

Salesforce acquires Listen Labs for ~$2b. But who gets the 💰? My usual breakdown below 👇 The company was founded in Sep-23 and had raised less than $100m. Investors will share ~$850m of profits on $96m invested, a ~10x blended. From a huge pivot to a unicorn valuation term sheet they refused, this is truly a wild story ✍️. In the end an incredible outcome for all involved. So let's dive in! 1) The boldest move: walking away from $1.5b 🎲 Listen Labs had a signed $125m Series C term sheet from Menlo at a $1.5b valuation... and walked away from it to sell to Salesforce at ~$2b. This takes a huge amount of courage: few founders turn down a signed unicorn-plus round. @itsalfredw and @florian_jue did it 2.5 years after founding. Huge congratulations are in order for the discipline and execution of the founders here. 2) Founders and team: a life-changing outcome in under 3 years 🥳 By my best estimates, founders and team still own over half the company. At $2b, that's ~$1.1b for them to share. Specifically, assuming a 15% option pool, that is ~$300m for the team and ~$750m for the two co-founders. And, as was the case with Hugging Face, this is all from a pivot. They originally built an AI customer-interview tool to understand why their viral app BeFake was growing, then realised the tool was the business! 3) Sequoia's Bryan Schreier did it again 👑 What few people know is that Bryan was an early backer of Qualtrics... the category Listen Labs is disrupting. Now he led both the seed and the Series A here. By my estimates, those two rounds will return ~$670m combined, or ~25x on ~$27m invested, in under three years. Pattern recognition and industry knowledge have their perks, it would seem! 4) Ribbit: 4x in 8 months ⚡ Ribbit led the $69m Series B in Jan-26 at ~$500m. At $2b, that's ~4x in 8 months. Unbelievable IRR and a great return on a meaningful cheque. 5) Neo does it again, congrats @apartovi 🎯 Listen Labs went through the Neo accelerator early on, which came with a $600k SAFE. On my estimates, that cheque is worth ~$25m+ today. This comes just three months after Cursor's acquisition (a cool >1,000x for Neo). What a hit rate! This one is very straightforward: it is a massive win for everyone. And so congratulations to all involved: Sequoia, Ribbit, Conviction, Pear, Neo and the team!

most winning poker players will never say this stuff publicly because it costs them games & future ev standing up for C Ye & telling this story about Britney takes massive balls I don’t think people realize how much courage this takes, respect.

C Ye @C_Ye__ & Britney’s Drama reminds me another story back in 2023. It’s kinda like Deja Vu. Britney leveraged her fame and connections to bully an unknown small-time poker player. I’ve seen many hustler players backing Britney. A lot of them were coerced into supporting her, yet none stepped up to speak up for C Ye. To me, this is grossly unfair and amounts to collective bullying. Most of them don’t even know what really happened. They may simply fear that if they fail to back her, she will get them banned from Hustler @HCLPokerShow since she is the “Queen” of hustler. XD This happened in 2023 at the Bicycle Casino in Los Angeles. Britney set up a game with Peter. She personally brought in two players, Hank and Jimmy, and asked another host, L, to bring five more. (L is just a codename — he is a very well-known Chinese host, but he prefers not to have his name mentioned here.) Most importantly, Britney bought action in all of the players — every player’s wins and losses ultimately ran through her. Charles the Prince also played in the game for about an hour and had action in Hank as well. That means both L and Charles were financially exposed to Hank’s play that night. To this day, L still believes that Hank and Britney were together at the time. The players on L’s side ended up losing. But when it came time to settle up, Britney refused to pay, claiming that someone had stolen chips from the table. The game itself balanced, though — every number was accounted for. Since Britney had action in all of the players, the “stolen chips” story looked less like an explanation and more like an excuse to avoid settling. And when people pushed back, she firmly denied owing the debt. Alan, an Australian regular who used to play at the Bike, happened to be there that night, and he took the matter into his own hands. He spent 36 hours on it. For the first 20-plus hours, he negotiated with the casino and obtained everyone’s deposit and withdrawal records along with the surveillance footage. The review showed that the person who had actually taken the chips was Hank — one of Britney’s own players, and reportedly someone very close to her at the time. Alan handed all of the evidence to Britney. She still refused to pay — and even after all of this came out, she kept up her good relationship with Hank. Only after their relationship later ended did she start distancing herself from him and publicly criticizing him, as if she had never had anything to do with him. Later, L — who was in Vegas at the time — called Britney and warned her that the police would be involved if the payment wasn’t made. Meanwhile, Alan and several other pros were right there at the Bike. Alan never left — he stayed the entire time for one purpose: to get his money. He was relentless. In the end, Britney had no choice but to pay up. Start to finish, the whole thing took 36 hours.

I’ve seen a couple of posts about this so wanted to demystify. Today, every Muse user gets a free computer in the cloud. It's a real computer, and we’ve designed the security architecture of the Muse Secure VM carefully so you and your Muse can do almost anything you could with a computer sitting under your desk while keeping you and the system safe from threats like prompt injection. We wrote about this at length in our security blog post – Activity in the “runtime cell”, which you share with your Muse is unfettered, but sensitive actions are all overseen by the Sentinel, which runs outside of that cell. Similarly, all sensitive secrets - like the passwords you enter into Muse’s secure credential storage - are also stored outside the runtime cell. The runtime cell gets its own root filesystem (including a full Ubuntu linux image) separate from the host filesystem where your other more sensitive data lives. Because it is isolated from the sensitive stuff that runs on the same box, this means that we can, and do, offer users full visibility and control over the files in the runtime cell. Just as you can when you install Linux on your home computer, you can poke around and see all the files that make the system work - both debian system files and the binaries and data files that implement the parts of Muse which run in the runtime cell. This was a very deliberate choice - your Muse Secure VM truly is your own computer in the cloud. You can install software in it, write and compile code, use the browser to surf the web: it is your own Linux box that you can operate as you choose with your Muse. Poking around in this computer doesn't give you any privileged access to Meta infrastructure, or to other people's data If I may geek out a little here for a second… As a kid I loved to take things apart to see how they worked. As a teenager I got into computers and soon found myself drawn to C:\WINDOWS\SYSTEM and the system registry, later Slackware’s /dev/, /proc/ etc – I could see how the system was laid out and as I explored what DLL files and .so files actually did, I gradually became able to meld the computer to my own will. We’re really proud to be able to put a real computer in millions of people’s hands with a similar level of transparency. We built a file explorer right into the Library tab of the UI. We want you to be able to see the markdown files Muse writes while it thinks about how to serve you better, and explore the internals of the system if you’d like to. So, when you ask your Muse to show you its entire filesystem, and receive gigabytes of files you’re seeing the full contents of the runtime cell. It’s yours to explore and enjoy! If you’re not a geek like me, or simply want to download the data that you personally have created directly with your Muse, we added a feature for that too in Settings > Data controls > Download your agent data.

Une annonce hors normes. Vous connaissez mon intérêt pour les mangas. Vous savez aussi ma détermination à attirer en France les investissements qui créent des emplois et rendent possibles les projets les plus ambitieux. Avec l’Arabie saoudite, nous faisons une annonce sans précédent. À Cergy-Pontoise, 3 parcs à thème vont voir le jour. 6 milliards d’euros d’investissements. 22 000 emplois créés. Du jamais vu depuis Disneyland Paris. Une nouvelle destination mondiale. Ici, en France. Merci à Qiddiya et à l’Arabie saoudite pour leur confiance en notre pays, en nos talents et en notre avenir. C’est tout le sens de Choose France : aller chercher les projets les plus ambitieux et faire de la France le pays où ils deviennent réalité. Créer des emplois. Faire rayonner la France. Nous n’avons pas fini de surprendre le monde. Très fier de cette réussite.

Following my last collaboration with 911, this time I teamed up with C.Holly! 💕 This is also my second song in Chinese 🎶 The tempo of this song is so fast, so recording it was really challenging 😂🤣 But C.Holly and I both worked incredibly hard to finish it!! So what I’m most curious about is… Can you understand my Chinese pronunciation? 🥹😂 The music video is finally out!! Go watch it now 💖🎬 C.Holly X DJ SODA – 花心大蘿蔔 🌸