Gemma 2 9B looks like a strong model for creative writing, especially with length correction. Here's a sample from EQ-Bench with the prompt, result, and analysis from Claude 3.5 Sonnet.
@maximelabonne
-
AI Engineer Conference Presentation Judging Highlight
By
–
Getting judged by @altryne
-as-a-Judge at @aiDotEngineer Funniest presentation of the conference, thanks for this gem -
Fine-tuning and Model Merging: Complete Overview Guide
By
–
I made a quick and dense overview of everything you need to know about fine-tuning and model merging at @aiDotEngineer I'm sharing here the video and my presentation slides. Thanks everyone for your feedback! Slides: https://
drive.google.com/file/d/1bb6zTF
qRIa_ai23fvxq4GFFMkoGZpJb7/view?usp=sharing
… Video: -
Fine-tuning Models and Creating Llama 3 120B Instruct
By
–
My talk at @aiDotEngineer is starting soon! Come to Salon 8 to learn when to fine-tune models and how to create monstrosities like Llama 3 120B Instruct.
-
Model Merging Preserves Safety Alignment Better
By
–
Model Merging and Safety Alignment New paper looking into how model merging poorly preserves safety alignment. The authors modify EvoMM and LM-Cocktail to balance performance on safety data and domain-specific data. They show that this safety-aware merging approach can
-
New Preference Alignment Technique Combines Model Merging and RLHF
By
–
We've seen a lot of new preference alignment techniques lately, but this is a really cool one combining model merging and RLHF.
-
The Hidden Complexity of Image Inpainting in AI
By
–
I just wanted to remove a button They never tell you that inpainting is hard
-
AI Safety Paper and Code Released on Refusal Direction
By
–
The authors of the original blog post released the paper and code! Paper: https://
arxiv.org/abs/2406.11717 Code: https://
github.com/andyrdt/refusa
l_direction
… -
Structured Generation Libraries: Comparing AI Frameworks
By
–
Libraries for structured generation Nice blog post by @sxlijin (also the creator of BAML) about the different frameworks to get structured generation I've personally tested JSONformer, Guidance, LMQL, and @OutlinesOSS (in this order). I would definitely recommend Outlines