Zaoral @zaoral@fedibird.com

https://techblog.ap-com.co.jp/entry/2024/05/26/163419

https://note.com/masayuki_abe/n/nc0e6d7f78397?sub_rt=share_pw&fbclid=IwZXh0bgNhZW0CMTEAAR2WUlf7BhyOyO5nOiojnOm7AeVV9_YIKwwme1kDgaQ4dD8qJmBnpNoy4LM_aem_AaNvsoraOVSbaOKwUa_5jIL3w4yuzdkquNbKu9U-a_a8l9nV9Nu5UeQriMYKe3Y-4LdVTl3WanOJmZVMabhFsdDg
llama3 scrach

https://twitter.com/ai_database/status/1793628905141883207
rag 失敗

https://note.com/npaka/n/nb050244392a4?sub_rt=share_h&fbclid=IwZXh0bgNhZW0CMTAAAR3aBo6mlhdcSU0FkTAMyj7Q-Ds0uMOG3cqhll9wWzcf1PGtPxQhO7ibF_Q_aem_AbL3GCfyEj1RpLScjmRPcLKXRsk3YVF4vkivoIu9rpAukByezjkxz0s-bImd-zFhsBDHZNZ1rKMgx2REr1obnl3_
phi3

https://twitter.com/arankomatsuzaki/status/1792746476260790457
MoRA
https://github.com/kongds/MoRA
https://arxiv.org/abs/2405.12130
>Outperforms LoRA on memoryintensive tasks and achieves comparable performance on other tasks

https://zenn.dev/knowledgesense/articles/346183afdf750e
Binary Embedding

Machine Unlearning
https://twitter.com/Maxwell_110/status/1793037804144054702

https://twitter.com/Maxwell_110/status/1792313028475781253
スクラッチLLM

https://huggingface.co/blog
Blog, Articles, and discussions

https://github.com/huggingface/transformers/releases/tag/v4.41.0
uggingface
/
transformers
v4.41.0:

https://twitter.com/shanegJP/status/1791616634550911187

https://twitter.com/WMjjRpISUEt2QZZ/status/1791475865580372079

https://twitter.com/CurveWeb/status/1790904543817125939

https://llm.stockmark.co.jp/
https://twitter.com/kosukearima/status/1790970299103809957
1000億パラメータの日本語LLMモデル「Stockmark-100b」

https://twitter.com/npaka123/status/1790455092203700317
Gemini 1.5 Pro

https://gigazine.net/news/20240514-falcon-2-ai-model/
Falcon 2

https://huggingface.co/tiiuae/falcon-11B

https://zenn.dev/acntechjp/articles/9041d574cf4f54?fbclid=IwZXh0bgNhZW0CMTEAAR2cicgG26vMKpln0UsVW4Jb_m4KCfJ39d1HYgifYASSzI5CuYC5wVqHuUM_aem_AdUx6TMUsMGIe815bZjMiSiwvMB5MbqUkn2PiRak12WWivK0TUTnAbBTutrSjOp5LoYZCWNLqXZuE3VNmkE_jKpu

https://www.deeplearning.ai/the-batch/how-agents-can-improve-llm-performance/
ng エージェント

https://twitter.com/nikkei/status/1790215644945797470

https://twitter.com/gyakuse/status/1790110045814010327

https://huggingface.co/spaces/sakasegawa/gpt-4o-tokenizer-vs-gpt-4-tokenizer
gpt-4o-tokenizer-vs-gpt-4-tokenizer