RT @aryopg: Is there a cheap way to adapt LLMs to the clinical domain? Totally! We employ Parameter-Efficient Fine-Tuning (PEFT) techniques, and we take it even further by doing it twice: one for domain-adaptation pretraining and another for task fine-tuning! 👉arxiv.org/abs/2307.03042 🧵1/6 pic.twitter.com/xPxMd6KorI
posted at 07:55:32
RT @GregKamradt: I demo'd a method to summarize an entire book w/o sending 100% of the tokens to an LLM. The results weren't bad! Then @musicaoriginal2 DM'd saying he was open to adding it to @LangChainAI. @RLanceMartin merged today. So cool to see this go full circle twitter.com/GregKamradt/st… pic.twitter.com/KRJHxEP2pr
posted at 07:53:43
RT @nelsonfliu: LM context windows are getting larger, but how well do they actually _use_ long input contexts? We find that LM performance is highest when relevant info occurs at the very start or end of long input contexts, and significantly degrades in the middle. arxiv.org/abs/2307.03172 🧵 pic.twitter.com/7oesT2YU3O
posted at 07:42:51
RT @roadjiang: Fascinating research by Google reveals the power of Language Models (LLMs) like PaLM or GPT in tackling visual tasks using in-context learning. This novel method enables LLMs to perform image generation tasks without requiring any parameter updates. #palm #GPT4 #LLMs pic.twitter.com/m2EiDj35rv
posted at 07:39:40
RT @DrJimFan: Interesting work from Google that tokenizes an image using LLM's native word tokens. Typically VQ-VAE/VQ-GAN methods learn their own codebook, which wouldn't be understood directly by blackbox LLMs. The tokens are arranged in a spatial pyramid. Deeper layers -> more refinements twitter.com/i/web/status/1… pic.twitter.com/nkT8U1MA6B twitter.com/roadjiang/stat…
posted at 07:38:57
RT @llama_index: Pretty much everyone building LLM apps over data has to figure out how to deal with the following: How to cram arbitrary data into limited context windows 📦 Our 0.7.0 response synthesis modules eliminate the need to write this boilerplate. Here’s an overview of strategies 👇 pic.twitter.com/0RSndukyl5
posted at 07:29:26
RT @techtalkjp: OpenAI API コール、これやったら日本語固定にできた。user よりもあとに system をぶっこむ。 messages: [ { role: 'system', content: systemPrompt }, { role: 'user', content: input }, { role: 'system', content: '出力は日本語です' }, ],
posted at 07:28:52
RT @nlp_colloquium: 🚀今後のNLPコロキウムのお知らせ 7月からも豪華なゲストに研究トークをしていただきます。お楽しみに！ @yu_takagi @michiyasunaga pic.twitter.com/OF2J9jJ9JA
posted at 00:14:49