Komorebi AI

KomorebiAI's activity

• All
• Models
• Datasets
• Spaces
• Papers
• Collections
• Community
• Posts
• Articles

vicgalle 

authored a paper about 1 month ago

Distilling Feedback into Memory-as-a-Tool

Paper • 2601.05960 • Published Jan 9 • 2

vicgalle 

submitted a paper to Daily Papers about 1 month ago

Distilling Feedback into Memory-as-a-Tool

Paper • 2601.05960 • Published Jan 9 • 2

vicgalle 

updated a collection 5 months ago

Multimodal Models 🔀

Collection

A collection of multimodal models developed by the Komorebi AI team • 3 items • Updated Sep 23, 2025 • 2

vicgalle 

authored a paper 7 months ago

Specification Self-Correction: Mitigating In-Context Reward Hacking Through Test-Time Refinement

Paper • 2507.18742 • Published Jul 24, 2025 • 6

vicgalle 

updated 6 models about 1 year ago

KomorebiAI/nllb-200-1.3B-ct2

Translation • Updated Dec 19, 2024 • 2 • 2

KomorebiAI/nllb-200-1.3B-float16-ct2

Translation • Updated Dec 19, 2024 • 2 • 1

KomorebiAI/nllb-200-1.3B-int8-ct2

Translation • Updated Dec 19, 2024 • 11 • 2

KomorebiAI/nllb-200-3.3B-ct2

Translation • Updated Dec 19, 2024 • 2 • 2

KomorebiAI/nllb-200-3.3B-float16-ct2

Translation • Updated Dec 19, 2024 • 3 • 3

KomorebiAI/nllb-200-3.3B-int8-ct2

Translation • Updated Dec 19, 2024 • 13 • 3

vicgalle 

authored a paper over 1 year ago

Merging Improves Self-Critique Against Jailbreak Attacks

Paper • 2406.07188 • Published Jun 11, 2024 • 4

vicgalle 

authored 2 papers almost 2 years ago

Configurable Safety Tuning of Language Models with Synthetic Preference Data

Paper • 2404.00495 • Published Mar 30, 2024 • 2

Refined Direct Preference Optimization with Synthetic Data for Behavioral Alignment of LLMs

Paper • 2402.08005 • Published Feb 12, 2024 • 1

vicgalle 

posted an update about 2 years ago

view post

Post

Can you merge models of different sizes? ⚗️

Well, yes, if the models are somewhat compatible. Here is an experiment I did. I wanted to merge two of the best performing models: mlabonne/NeuralBeagle14-7B and jeonsworld/CarbonVillain-en-10.7B-v4

Here is my recipe:
1. Expand the layers of NeuralBeagle to 10.7B ala frankenmerge.
2. DPO-tune the previous model with a high-quality preference dataset, argilla/distilabel-intel-orca-dpo-pairs
3. Merge the previous model with CarbonVillain (needs —allow-crimes in mergekit! 🔪)

And here is the resulting model, CarbonBeagle-11B, which ranked top in the leaderboard for its size class:
vicgalle/CarbonBeagle-11B

• 

• 
• 
• 2 replies

·

🤝

15

15

🤯

11

11

+

Reply

vicgalle 

authored a paper about 2 years ago

Distilled Self-Critique of LLMs with Synthetic Data: a Bayesian Perspective

Paper • 2312.01957 • Published Dec 4, 2023 • 1

vicgalle 

authored 3 papers over 2 years ago

Fast Adaptation with Bradley-Terry Preference Models in Text-To-Image Classification and Generation

Paper • 2308.07929 • Published Jul 15, 2023 • 1

Personalizing Text-to-Image Generation via Aesthetic Gradients

Paper • 2209.12330 • Published Sep 25, 2022 • 1

ZYN: Zero-Shot Reward Models with Yes-No Questions

Paper • 2308.06385 • Published Aug 11, 2023 • 1