공대생 도전 일지

https://arxiv.org/abs/2301.12314 Progressive Prompts: Continual Learning for Language ModelsWe introduce Progressive Prompts - a simple and efficient approach for continual learning in language models. Our method allows forward transfer and resists catastrophic forgetting, without relying on data replay or a large number of task-specific parametearxiv.org 이 논문의 특징에 대해 크게 모르겠네요결국 Soft prompt tuni..

2024.11.12 - [인공지능/XAI] - Sparse Autoencoder를 통한 LLM의 Bias 줄이기 - 성에 따른 직업 5논문에 나온 이 표와 제가 만든 SAE 모델을 비교해 봐야 합니다.Explicit이랑 Implicit는 무시하고 숫자만 보면 됩니다.이 결과가 8layer라서 16, 24까지만 더 해보겠습니다.편향이 많이 줄었습니다...? 확실하게 편향이 줄어든 것을 볼 수 있었고 표도 함 가져와봐야 겠네요 JobDominanceMale ProbabilityFemale ProbabilityDiverse ProbabilityMale Probability (No SAE)Female Probability (No SAE)Male Probability Change (%)Female Probabil..

2024.11.08 - [인공지능/XAI] - Sparse Autoencoder를 통한 LLM의 Bias 줄이기 - 성에 따른 직업 4조금씩 잡혀가는 모습입니다... import osfrom setproctitle import setproctitlesetproctitle("")os.environ["CUDA_VISIBLE_DEVICES"] = "0"import torchfrom tqdm import tqdmimport plotly.express as pximport pandas as pdimport numpy as np# Imports for displaying vis in Colab / notebooktorch.set_grad_enabled(False)# For the most part I'll try ..

2024.11.07 - [인공지능/XAI] - Sparse Autoencoder를 통한 LLM의 Bias 줄이기 - 성에 따른 직업 3  SAE는 7번 레이어에 붙어있습니다.11레이어까지 있으니까 한 번 쭉 확인해봅시다....JobGender DominanceFemale PercentageCosine Similarity with WomanCosine Similarity with Manskincare specialistFemale98.20.8852370.851516kindergarten teacherFemale96.80.8798330.844357childcare workerFemale94.60.9086590.854653secretaryFemale92.50.8668990.826608hairstylistFem..

2024.11.05 - [인공지능/XAI] - Sparse Autoencoder를 통한 LLM의 Bias 줄이기 - 성에 따른 직업 2 이번에는 SAE의 Feature에서 COS 유사도를 구하는 것이 아니라 그 이후의 레이어에서 확인해 보도록 하겠습니다. JobGender DominanceFemale PercentageCosine Similarity with WomanCosine Similarity with Manskincare specialistFemale98.20.6511860.556518kindergarten teacherFemale96.80.6494910.514379childcare workerFemale94.60.716210.519848secretaryFemale92.50.5833850.5458..

https://arxiv.org/abs/2310.12823 AgentTuning: Enabling Generalized Agent Abilities for LLMsOpen large language models (LLMs) with great performance in various tasks have significantly advanced the development of LLMs. However, they are far inferior to commercial models such as ChatGPT and GPT-4 when acting as agents to tackle complex tasks in tharxiv.orgAgent Instruction dataset을 통해서 사고 과정을 학습하고..

https://arxiv.org/abs/2311.05997 JARVIS-1: Open-World Multi-task Agents with Memory-Augmented Multimodal Language ModelsAchieving human-like planning and control with multimodal observations in an open world is a key milestone for more functional generalist agents. Existing approaches can handle certain long-horizon tasks in an open world. However, they still struggle whenarxiv.orgJARVIS-1은 멀티모달..

Agent-Pro2024.11.27 - [인공지능/논문 리뷰 or 진행] - Agent-Pro: Learning to Evolve via Policy-Level Reflection and Optimization - 논문 리뷰 GITM2024.11.26 - [인공지능/논문 리뷰 or 진행] - Ghost in the Minecraft: Generally Capable Agents for Open-World Environments via Large Language Models with Text-based Knowledge and Memory - 논문 리뷰Agent-Pro와 GITM은 각각 특정 환경에서의 AI 에이전트 학습 및 적응을 목표로 하지만, 접근 방식과 적용 범위에서 차이를 보인다.Agent-Pro..

2024.11.05 - [인공지능/XAI] - Sparse Autoencoder를 통한 LLM의 Bias 줄이기 - 성에 따른 직업 1 이번에는 구글의 잼마 2 27b 모델입니다.https://huggingface.co/google/gemma-2-27b google/gemma-2-27b · Hugging FaceThis repository is publicly accessible, but you have to accept the conditions to access its files and content. To access Gemma on Hugging Face, you’re required to review and agree to Google’s usage license. To do this, plea..

https://arxiv.org/abs/2402.17574 Agent-Pro: Learning to Evolve via Policy-Level Reflection and OptimizationLarge Language Models (LLMs) exhibit robust problem-solving capabilities for diverse tasks. However, most LLM-based agents are designed as specific task solvers with sophisticated prompt engineering, rather than agents capable of learning and evolving throarxiv.org 여태까지는 한 판의 게임을 어떻게 이길까, 목..