Zaoral @zaoral@fedibird.com

https://zenn.dev/aratako_lm/articles/a5ae43fb2bfbb3?fbclid=IwY2xjawFvm39leHRuA2FlbQIxMQABHXToT2OaYzHVLm6YQEESQXZ1czjTIIQ3N91LqDm81RRv4V1JywurxykA7A_aem__JYqUxSdi8b9rb3FfRfeEg

https://x.com/sleepy_yoshi/status/1841880900155646202
https://t.co/IYthINas1E
Mistral-NeMo-Minitron-8B-Instruct

https://huggingface.co/collections/google/gemma-2-jpn-release-66f5d3337fdf061dff76a4f1

https://x.com/rinna_research/status/1841670544674193838

https://huggingface.co/spaces/alfredplpl/uno-gemma-2-2b-jpn-it

https://github.com/unslothai/unsloth

https://x.com/izutorishima/status/1841454788749484378

https://x.com/novelaiofficial/status/1841043853932933302
Llama 3 Erato

https://blog.novelai.net/クイックスタートガイド-日本語-9bd63ac4be72

日本語Qwen2.5であるEZO-Qwenの72B版がGGUF化
https://x.com/2022_technology/status/1840354395906834619

https://huggingface.co/grapevine-AI/EZO-Qwen2.5-72B-Instruct-GGUF

https://x.com/bwgift20th/status/1839861830283403365

https://x.com/jaguring1/status/1839047872039497864

Llama 3.2

https://medium.com/alan/llama-3-takeaways-for-llm-practitioners-209c50e3892f

https://x.com/ai_scholar/status/1838499164910674348

https://ai-scholar.tech/articles/large-language-models/Kolmogorov-Arnold-Network

https://arxiv.org/abs/2302.05019

LLM-jp-3 1.8B
https://x.com/npaka123/status/1838835688541950256

https://x.com/jaguring1/status/1838615823298236489

Gemini 1.5 pro 002
Gemini 1.5 Flash 002

https://x.com/schroneko/status/1838192580519157911

https://huggingface.co/datasets/openai/MMMLU

> OpenAI の多言語性能評価ベンチマーク MMMLU (Multilingual Massive Multitask Language Understanding)

https://x.com/TetsuroMorimura/status/1837302545364263031

https://arxiv.org/abs/2404.13846

>LLMアライメント（RLHF）の標準となりつつあるDPOは、学習データに質の低い応答文が含まれると、従来のRLHF以上に性能が劣化

https://huggingface.co/ModelCloud/GRIN-MoE-gptq-4bit
GRIN-MoE

https://x.com/jaguring1/status/1836457287566921752