Yes, for sure. Distillation would be ideal but DPO would also work.
MACHINE LEARNING
-
Google Showcases 70+ Papers at ECCV 2024 Milan
By
–
This year Milan, Italy hosts #ECCV2024, which kicks off today with a full docket of workshops and tutorials. Google has a strong presence at @eccvconf with over 70 accepted papers and participation in over 25 workshops and tutorials. Find the full list @ https://
goo.gle/eccv24 -
Self-Merge Experiments: 52B Model Success in BBH and MATH
By
–
I want to share some results related to my self-merge experiments: – The 47B merge doesn't work: replicating blocks > layers (apologies for doubting it) – The 52B is successful in BBH and MATH Lvl 5. However, there might be an issue with the original model's MATH eval. This is
-
Oryx MLLM: Spatial-Temporal Understanding at Arbitrary Resolution
By
–
Oryx MLLM: On-Demand Spatial-Temporal Understanding at Arbitrary Resolution https://
oryx-mllm.github.io https://
arxiv.org/abs/2409.12961
The work proposes Oryx, a unified multimodal architecture for the spatial-temporal understanding of images, videos, and multi-view 3D scenes. -
Critic Plasticity Loss Bottleneck in Deep Reinforcement Learning
By
–
(1) data augmentation is essential in maintaining plasticity; (2) the critic's plasticity loss serves as the principal bottleneck impeding efficient training; and (3) without timely intervention to recover critic's plasticity in the early stages, its loss becomes catastrophic.
-
Plasticity in Visual Reinforcement Learning: Data and Training
By
–
Revisiting Plasticity in Visual Reinforcement Learning: Data, Modules and Training Stages #ICLR2024 https://
arxiv.org/abs/2310.07418 https://
github.com/Guozheng-Ma/Ad
aptive-Replay-Ratio
…
we conduct a systematic empirical exploration focusing on three primary underexplored facets and derive the following conclusions: -
Real-time Object Detection with YOLOv5n6 for High-Resolution CCTV
By
–
CCTV_YOLO Fast Real-time Object Detection with High-Res Output This project demonstrates real-time object detection using the YOLOv5n6 model with low-resolution inference for high-speed processing, while drawing the results on high-resolution frames. The object detection
-
HARP: ML-Powered Audio Sample Editor with Gradio Integration
By
–
HARP HARP is a sample editor that allows for hosted, asynchronous, remote processing of audio with machine learning. HARP works by routing audio through Gradio endpoints. Since Gradio applications can be hosted locally or in the cloud (e.g. with HuggingFace Spaces), HARP lets
-
RL Applications in Chip and PCB Hardware Engineering
By
–
Thanks, @Kbeguir
! The application of RL to chip floor planning (AlphaChip and follow on work), and @instadeepai
's work on RL for PCB board design (DeepPCB) are both great examples of real world application of RL to hardware engineering problems! -
Distributed Neural Network Training: From 1990 to Large Scale
By
–
I did an undergrad thesis on parallel training of neural networks in 1990, then detoured through HIV/AIDS forecasting, compiler optimizations, information retrieval, distributed computation and storage systems before returning to large scale distributed training of neural