Nico Kreiling
234 posts
Nico Kreiling
@NicoKreiling
coder & tech enthusiast from Cologne, data scientist @scieneers_gmbh and host of https://t.co/lbAyO9KOOO
Köln, Deutschland Katılım Mayıs 2015
207 Takip Edilen254 Takipçiler
📅 Next week, I have the opportunity to speak at the #itdays about personalization using #generativeai
🤖 To support this talk, we used #chatGPT to build an agent, that recommends the best talks for every time slot based on your interests: share.streamlit.io/app/2023-it-da…
English
Nico Kreiling retweetledi
📣 Introducing ⭐ StarCoder+ & StarChat Beta!
We trained StarCoder on the Falcon model's English web dataset and Instruction-tuned it. Both models rank high in the LLM leaderboard, with strong natural language performance and coding capabilities.
huggingface.co/HuggingFaceH4/…
English
@ClementDelangue Totally agree. I don't tweet that much, but my thread in falcon didn't even get one like 🙈 twitter.com/nicokreiling/s…
Nico Kreiling@NicoKreiling
The Falcon model recently outperformed Llama according to the LLM leaderboard and it's license recently got changed to Apache 2. But did not everyone say, only big tech companies can train such a model? So who are the others and how was it financed? 🧵
English
IMHO: It's nice, that Abu Dhabi invests its oil millions into well-needed open-source technology innovations!
English
If this were correct, Falcon would have been a bit cheaper in training compared to other LLMs (50% GPT3). However, even if it was more expensive or they needed multiple runs to get it right, this should be doable with Abu Dhabi on your back.
English
The Falcon model recently outperformed Llama according to the LLM leaderboard and it's license recently got changed to Apache 2.
But did not everyone say, only big tech companies can train such a model? So who are the others and how was it financed? 🧵
English
@programmierbar Heyho :) Ich bin ja auch fleißiger Programmier.bar Hörer und dachte, ich verlink euch mal noch meine Techtiefen Folge die ich gerade mit Andreas von OpenAssistant aufgenommen habe: techtiefen.de/42-ai-chatbots…
Deutsch
Neues aus der Tech-Welt:
👉 Das bringt #TypeScript 5.1
👉 Deno für next level Datenbanken
👉 Beef zwischen Microsoft & Twitter
👉 OpenAssistant, neues chatbasiertes #opensource LLM
👉 StableLM, #LLM für kommerzielle Nutzung
Höre mehr in der neuen Folge: programmier.bar/podcast/news-1…
Deutsch
Folge 42 #Techtiefen hat ein würdiges Thema: Es geht endlich um #chatgpt bzw. die open-source Alternative #OpenAssistant. Und neben den gewohnten Grundlagen auch mit der notwendigen Diskussion zu Ethik und gesellschaftlichen Auswirkungen.
Andreas Köpf@neurosp1ke
🫣German long-form 2.5h 🎙️techtiefen podcast (@NicoKreiling) interview with me about OpenAssistant; covers a lot of ground including doomsday AGI & social implications… I especially recommend listening the 2nd half or last half hour. techtiefen.de/42-ai-chatbots…
Deutsch
@NicoKreiling Sehr schön :) Im Halbschlaf oder im Auto war der Unterschied zu deutlich 😁
Deutsch
@NicoKreiling Bitte die Podcast-Folge 42 normalisieren und reuploaden. Dein Gegenüber ist sehr leise
Deutsch
A small recap of model sizes for myself (and an hopefully upcoming Techtiefen episode).
English
Let's see if StableLM as a competitive and really open (also for commercial use) new base model will bring such an explosion in creative use cases as StableDiffusion did.
Stability AI@StabilityAI
Announcing StableLM❗ We’re releasing the first of our large language models, starting with 3B and 7B param models, with 15-65B to follow. Our LLMs are released under CC BY-SA license. We’re also releasing RLHF-tuned models for research use. Read more→ stability.ai/blog/stability…
English
Just learned about learned DBMS by @cbinnig, and that ML models can much better predict query runtimes, but training them is to much effort. Maybe a similar method as TabFPN might work here?
Frank Hutter@FrankRHutter
This may revolutionize data science: we introduce TabPFN, a new tabular data classification method that takes 1 second & yields SOTA performance (better than hyperparameter-optimized gradient boosting in 1h). Current limits: up to 1k data points, 100 features, 10 classes. 🧵1/6
English
Excited to see talks about what people are building with Haystack @PyConDE! 🥳
@NicoKreiling and Alina just presented how they created a Slack QA bot using #Haystack 🤩
Berlin, Germany 🇩🇪 English
@RitchieVink Yes I did, but I wasn't the only one, actually there have been two talks about polars in a row 😅 slides are here dropbox.com/s/fphay3yav2b2…
English
@NicoKreiling Wow! Awesome to see you gave a talk about polars!
How was the reception?
Are the slides available?
English
So great to speak at a PyCon for the first time 😍
And 🧡the community. After my talk, someone pointed me to an issue I had overseen before. Feel free to promote it if you also want to use #polars for EDAs: github.com/pola-rs/polars…
English