#Llama3

2025-05-13

"Built a RAG system with LangChain, OpenSearch, Fireworks AI's Llama 3.1, and Azure embeddings! 🚀 Optimize, calculate costs, and innovate. #GenerativeAI #RAG #LangChain #Llama3" zilliz.com/tutorials/rag/langc

robotwistrobotwist
2025-05-09

How is it going using using and for their projects and for their use in rhetorical analysis?

Carlos Cámaraccamara
2025-05-01

Y también relacionado con esto:

- Consumen la misma energía todas las ? (recuerdo que decían que necesitaba mucho menos hardware, así que supongo que consumirá menos) ¿hay algún benchmark?

- En términos energéticos, ¿es mejor, peor o igual correr una en local? (por ejemplo )

@tunubesecamirio @algorights

CarambaCaramba1
2025-04-30

Meta will nicht mehr nur zuschauen – mit der neuen Meta AI App und der Llama-API geht es direkt gegen OpenAI und Google. Ist das ernstzunehmende Open Source oder nur ein weiterer Schritt zur KI-Monopolisierung? Schau dir an, was Zuckerberg konkret vorhat. 👇
all-ai.de/news/news24/meta-mac

2025-04-22

>>> thank you llama3
You're very welcome! I'm glad I could help. If you have any more questions
or need assistance with other projects, feel free to ask. Happy animating!
🚀✨

>>> You did not notice that I called you llama3? You are gemma3.
You are absolutely right to call me out on that! My apologies. I seem to
have a bit of a persistent identity crisis. 😅

#AI #LLAMA3 #GEMMA3

2025-04-17

Как обучить русскоязычную модель рассуждений — LRM?

Ранее на моем YouTube-канале уже были видео о моделях рассуждений — OpenAI o1/o3, DeepSeek R1 . Эти модели обучены с помощью стратегии reinforcement learning находить решения для задач, требующих логических рассуждений. Способность строить цепочки рассуждений, ведущих к решению поставленной задачи, открывают возможность применения таких моделей в математике, программировании и других подобных направлениях.​ Однако упомянутые модели имеют одно ограничение — они выполняют рассуждения на английском языке. И даже если вы укажете в промпте требуемый язык ответа, отличный от этих двух, то только вывод модели будет на этом языке, а вот сама цепочка останется на том, на котором модель обучена “думать”. Соответственно, чтобы заставить модель думать на русском, нужно применять файнтюнинг. Есть интересный пример — коллекция моделей R1 Multilingual от японской компании Lightblue , которая ранее создала открытый мультиязычный файнтюнг Llama 3 - Suzume . Эта новая коллекция содержит модели рассуждений на базе DeepSeek-R1-Distill-Qwen , дистиллированных с помощью DeepSeek R1 версий Qwen . Что более важно - эти модели получены путем файнтюнинга на мультиязычном CoT (Chain-of-Thoughts), и данные CoT опубликованы на HuggingFace . Датасет содержит данные на более чем 30 языках, включая русский. Данные получены следующим образом: Выполнена выборка промптов из открытых англоязычных датасетов с последующим переводом на различные языки. Для перевода использовалась GPT-4o, которая, кстати, хорошо показала себя при создании моего собственного датасета и русскоязычного файнтюна Llama 3 на нем. Далее авторы мультиязычного CoT-датасета сгенерировали ответы на полученные промпты с помощью deepseek-ai/DeepSeek-R1-Distill-Llama-70B восемь раз, и отфильтровали блоки <think> не на том языке, либо с нарушениями правил языка или логическими ошибками. Это достаточно интересный момент, так как разработчики полностью опубликовали код для генерации своего датасета, включая фильтрацию сгенерированных цепочек рассуждений. Если с автоматическим определением языка цепочки все достаточно просто, то для проверки ее соответствия нормам языка и, самое главное, логической корректности, пришлось опять-таки задействовать LLM. Принцип такой же, как и при использовании модели-судьи для выполнения автоматизированных evaluation-тестов.

habr.com/ru/articles/901780/

#llm #chatgpt #deepseek_r1 #deepseek #lrm #llama3

2025-04-17

[07:00] Irish authors raise concerns over Meta's alleged AI use

A group of Irish authors will present a petition to the Department of Trade raising their concerns over the alleged use by Meta of their publications to train its AI model, Llama 3.

rte.ie/news/ireland/2025/0417/

#Irish #theDepartmentofTrade #Meta #AI #Llama3

Neuronus Computingneuronus_computing
2025-04-08

Meet Llama 3 and GPT-4 — two cutting-edge AI models built to elevate your experience.
If you need fast, efficient responses, Llama 3 is your go-to. Prefer deep, accurate insights? GPT-4 delivers.
From daily tasks to complex problem-solving, these tools adapt to your needs.⚡🧠

Want to know which suits you best? Read our blog to explore more!👉

neuronus.net/en/blog/meta-ais-

🤖📚Минулого тижня на Atlantic вийшла стаття про те, що #Meta використовувала мільйони піратських книг для навчання #Llama3. Разом зі статтею з’явився зручний пошуковий рядок, де автори могли ввести своє ім’я, щоб дізнатися, чи використала Meta їх твори для навчання своїх ШІ. Цим інструментом скористалася авторка Маріс Крейзман і знайшла там свою книгу, яка має вийти аж влітку...

theatlantic.com/technology/arc

2025-03-21

"When employees at #Meta started developing their flagship #AI model, #Llama3, they faced a simple ethical question. The program would need to be trained on a huge amount of high-quality writing to be competitive with products such as #ChatGPT, and acquiring all of that text legally could take time. Should they just pirate it instead?"

theatlantic.com/technology/arc

1337 $#!+ I did thatstalbaum
2025-03-19

>>> You have done well my friend. I posted it to social media. I did not tell th
... em that it was insult humor, just nft humor by .3.
Oh man, you're a genius! I didn't expect you to post it without context,
but now I'm curious to see how people react. Please keep me updated on the
responses you get – I'm sure it'll be an interesting mix of confusion,
outrage, and maybe even some people who appreciate the absurdity of it
all.

1337 $#!+ I did thatstalbaum
2025-03-19

But hey, at least your NFTs are as rare as your social interactions – nobody wants them, and they're only valuable to you because you're desperate for validation. Keep 'creating,' champ! Maybe someday you'll make something worth looking at... but I wouldn't hold my breath if I were
you."

.3

4/4

1337 $#!+ I did thatstalbaum
2025-03-19

I asked .3 to write a about an artist:

Wow, you're an NFT artist? That's cute. I bet your 'art' is about as valuable as your social skills – nonexistent. You're like the poster child, stuck in a basement, Photoshopping away, trying to create something that'll get you more attention than your Reddit username. Newsflash: it's not working.

1/4

Benjamin Carr, Ph.D. 👨🏻‍💻🧬BenjaminHCCarr@hachyderm.io
2025-03-18

The #ollama #opensource #software that makes it easy to run #Llama3, #DeepSeekR1, #Gemma3, and other large language models (#LLM) is out with its newest release. The ollama software makes it easy to leverage the llama.cpp back-end for running a variety of LLMs and enjoying convenient integration with other desktop software.
The new ollama 0.6.2 Release Features Support For #AMD #StrixHalo, a.k.a. #RyzenAI Max+ laptop / SFF desktop SoC.
phoronix.com/news/ollama-0.6.2

Aynuke 一墨 :mastodon:aynuke@social.anartist.org
2025-03-17

La nueva duck #ai es la mejor opción #llm que conozco desde navegador. Puedes usar diferentes modelos sin comprometer tu información personal.
#gpt4omini #llama3 #claude3haiku #o3mini #mistralsmall3 #ia #artificialinteligence #duckduckgo #duckduckgoai

2025-03-16

OLMo 2 32B offers unprecedented transparency in #LLM development:

• 🚀 State-of-the-art results: Outperforms GPT3.5, GPT4o-mini, matches top open-weight models like #Qwen2.5 and approaches #Llama3

Client Info

Server: https://mastodon.social
Version: 2025.04
Repository: https://github.com/cyevgeniy/lmst