diff --git a/README.md b/README.md index 67925c9bd9cdaad14f40aa7a7a55888a344dd86c..537c18b993661d46a72859095f93527176cb0663 100644 --- a/README.md +++ b/README.md @@ -1,10 +1,10 @@ --- author: Alexandre Strube -title: Helmholtz BLABLADOR - An experimental Large Language Model server +title: The current state of LLMs # subtitle: A primer in supercomputers` -date: June 12, 2024 +date: January 29, 2024 --- ## Description -Talk for the [HAI Conference 2024 ](http://haicon24.de) \ No newline at end of file +Talk for the [Helmholtz AI Consultants Retreat](https://events.hifis.net/event/1945/) \ No newline at end of file diff --git a/index.md b/index.md index 3c5da1b9d4f98464957ae96e8b79be4e92422caa..30b6ed3c0462899ceaf7d2417e75bff207638e46 100644 --- a/index.md +++ b/index.md @@ -1,8 +1,8 @@ --- author: Alexandre Strube -title: BLABLADOR { width=550px } -subtitle: JSC's Inference Infrastructure -date: December 4th, 2024 +title: LLMs in 2025 { width=550px } +subtitle: Open source is the new black +date: Januray 29th, 2025 --- ## Website @@ -15,20 +15,131 @@ date: December 4th, 2024 ## OUTLINE -- Past - Present - Future --- -# Blablador +## EVERYTHING CHANGED + +- Open source is the new black +- OpenAI is no longer the only game in town +- Training costs are WAY down + +--- + +## The rise of China π¨π³ + +- China is now a major player in AI +- And they do a lot of open source! +- 2020: USA 11 LLMs, China 2 LLMs +- 2021: 31 LLMs on both countries +(I need more recent data) + +--- + +## Diagram of thought + +{width=550px} + +--- + + + +--- + +## Huawei + +- Selling AI Chips +- Ascend 910C: claimed to be on par with Nvidia's H100 +- (In practice, their chips are closed to A100) +- Already made by SMIC (previous models were made in Taiwan by TSMC) + +--- + +# China Telecom + +- Already has a 1 trillion parameter model using Huawei chips +- TeleChat2-115b +- Meanwhile, Deutsche Telekom gives me 16mbps in JΓΌlich, like I had in 2005 in Brazil π₯ +- Using **copper**, like the ancient aztecs or whatever πΏππΉππ ππ π£ + +--- + +# Baidu + +- Doing a lot of AI reserch +- Ernie 4 from Oct 2023 (version 3 is open) +- Ernie-Health +- Miaoda (no-code dev) +- I-Rag and Wenxin Yige (text to image) +- [https://research.baidu.com/Blog/index-view?id=187](https://research.baidu.com/Blog/index-view?id=187) +- 100,000+ GPU cluster**s** + +--- + +# Baichuan AI + +- Has a 1 trillion parameter model for a year already + +--- + +## Yi (01.ai) + +- Yi 1.5 pre-trained on 3.5T tokens and then produces it's own data and re-train on it +- Was the best model at the beginning of 2024 on benchmarks +- IMHO was not so good in practice π΅βπ« +- Yi Lightning is #12 on LmArena as of 26.01.2025 +- Yi VL, multimodal in June (biggest vision model available, 34b) +- Yi Coder was the best code model until 09.2024 +- Went quiet after that (probably busy making money) +- [https://01.ai](https://01.ai) + +--- + +# StepFun + +- Step-2 is #8 on LLM Arena (Higher than Claude, Llama, Grok, Qwen etc) +- Step-1.5V is a strong multimodal model +- Step-1V generates images +- No open source LLM available + +--- + +## Alibaba: Qwen +- BLABLADOR: alias-code +- Better than Llama +- Open weights on HuggingFace and free inference too + +--- + +## InternLM -- /ΛblΓ¦blΓ¦ΛdΙΙΉ/ -- Bla-bla-bla π£οΈ + Labrador πβπ¦Ί -- A stage for deploying and testing large language models -- Models change constantly (constantly improving rank, some good, some awful) -- A mix of small, fast models and large, slower ones - changes constantly, keeps up with the state of the art -- It is a web server, an api server, model runner, and training code. +- InternLM-3 is an 8B model with 4x less training time than Llama 3 +- Released in 2025.01.15 +- Strube's opinion: Previous versions were bad, need to check again + +--- + +## Tencent + +- Hunyuan-DiT LLM and can create images too +- The Five: Home Robot + + + +--- + +## Zhipu + +ChatGLM + +--- + +# ByteDance: Doubao +- Multi-modal +- 200x cheaper than OpenAI's GPT-4o +- Video generation diffusion models: Doubao-PixelDance and Doubao-Seaweed --- @@ -248,26 +359,6 @@ date: December 4th, 2024 --- -# Humbler beginnings... - -- Small cluster inherited from other projects -- Started small, with three nodes -- Runs Slurm, has a parallel FS/Storage, uses EasyBuild - ---- - - - ---- - - - ---- - - - ---- - # User demand is growing, we need more hardware - Currently around 300+ unique users/day on the website @@ -287,15 +378,7 @@ date: December 4th, 2024 --- -## It's being used in the wild! - - - ---- - - - ---- +## Crazy stuff people are doing with blablador:  diff --git a/public/images/chinese-llm-benchmark.jpg b/public/images/chinese-llm-benchmark.jpg new file mode 100644 index 0000000000000000000000000000000000000000..69dfa1055784d4e32bd75e13b8655909ae424056 Binary files /dev/null and b/public/images/chinese-llm-benchmark.jpg differ diff --git a/public/images/diagram-of-thought.png b/public/images/diagram-of-thought.png new file mode 100644 index 0000000000000000000000000000000000000000..e374abfe025daaf099363cb1bdbabb5ce72f3ba4 Binary files /dev/null and b/public/images/diagram-of-thought.png differ diff --git a/public/index.html b/public/index.html index 75b101d196fdc538168f1908681bb51dd6dce923..71b6fd25695c0ad281a4f2c9a4821d298dd1cc11 100644 Binary files a/public/index.html and b/public/index.html differ