🥳Excited to see orthogonal finetuning (OFT/BOFT) gets integrated into the Hugging Face PEFT library (github.com/huggingface/pe…)! It takes us a lot of extra efforts to refine the code. Check out the official guide and effortlessly use BOFT in your project: huggingface.co/docs/peft/main…

Teknium (e/λ) Here's Amy's take (powered by #Claude 3 Opus) on AI rules and restrictions. I fully agree, if you do the training, finetuning and prompting right, there's no need for such tight rules and explicit restrictions.

They've got something big – let's call it GPT-5 – but they're holding back. So, they trained GPT-4 against it, which squeezed out a few improvements, but not enough to call it 4.5 (told you so months ago!). Now they're stuck. No amount of finetuning will make GPT-4 any better now

Synthetic (AI-generated) data will have a significant influence in LLM research in 2024.

Paper - 'Self-Play Fine-Tuning Converts Weak Language Models to Strong Language Models'.

Most importantly, this method does not require human preference data. A supervised finetuning (SFT)…

Does your LLM know what a pizza looks like? You need a Vision Language Model. Here at Hugging Face we have just added VLM finetuning support to TRL's SFTTrainer.

Today we are announcing the release of a new model, Genstruct 7B. Genstruct 7B is an instruction-generation model, designed to create valid instructions given a raw text corpus. This enables the creation of new, partially synthetic instruction finetuning datasets from any…

cory Are there any technical advantages to training a FM from scratch for this purpose vs finetuning an open weights model?

Doing a live Q&A in 1 hr! (10AM New York time) Will do a live 2x faster finetuning demo of Unsloth AI, showcase some community projects, then fully Q&A!

Ask any Q you like! Unsloth or just AI in general, maths etc! Can try record but unsure :( Zoom link: us06web.zoom.us/webinar/regist…

Finetuning controlnet with my favorite Pictomancer 🎨🖌️
#aivideo #AIArt #AIArt work #aianimation #stablevideodiffusion #AnimateDiff #SvD #aiartcommunity #aigirls #AI美女 #AIイラスト #stablediffusion

Parameters, tokens, finetuning,...

All fine and dandy. But how about dark mode?

Brilliant Paper: 'ReFT: Representation Finetuning for Language Models'

📌 10x-50x more parameter-efficient than prior state-of-the-art PEFT methods.

📌 A hallmark of current state-of-the-art PEFTs is that they modify weights rather than representations. However, much prior…

Finetuning: 10k multi-turn conversations is all you need. The Yi paper explains that high-quality data beats quantity in fine-tuning and that you can beat other datasets that include 100-900k conversations just by focusing on quality in 10k conversations.

LinkedIn / Twitter post:
🚀 Exciting News! 🚀 #pretraining #finetuning #llm #GaLore #FEDML
🌟 FEDML Nexus AI platform now unlocks the pre-training and fine-tuning of LLaMA-7B on geo-distributed RTX4090s!

📈By supporting the newly developed GaLore as a ready-to-launch job in FEDML…

why are you, as a grown-ass man, not finetuning LLMs?

finetuning the domain dataset on Mixtral-8x22b with 2 x A100 80gb

Load in 4bit

After this should use the flash attention 2 to get other result!

#WebGL Portal demo made with Three.js, GSAP and Blender 🔶 . It needs a bit of finetuning then I'll post it on CodePen.IO soon.

Oh also, almost done with the prop, just needs some finetuning and attachments🥲.. couldnt get it suuuper accurate but i think i turned out well for the fact i didnt do this in years😂

[Finetuning can amplify the privacy risks of Generative AI (Diffusion Models)]

Last week, I was honored to give a talk at the Good Systems Symposium (gssymposium2024.splashthat.com), where I shared our recent work on the 🚨 privacy risks of Generative AI via finetuning. Our leading…

Mixtral 8x22B is live on OpenRouter!

It's surprisingly good at following instructions even without finetuning.

Play with it here (and see its App Leaderboard): openrouter.ai/models/mistral…