yjiangcm / lte Goto Github PK
View Code? Open in Web Editor NEWCode for "Learning to Edit: Aligning LLMs with Knowledge Editing (ACL 2024)"
Home Page: https://arxiv.org/abs/2402.11905
License: Apache License 2.0
Code for "Learning to Edit: Aligning LLMs with Knowledge Editing (ACL 2024)"
Home Page: https://arxiv.org/abs/2402.11905
License: Apache License 2.0
Excellent work! But I ran into OOM issue when i runing the lora method for llama-2-7b-chat model:
Specifically, I applied 4xA100(80GB) in the slurm system and then run the code below:
bash FastChat/lora_train.sh
The following error messages are displayed
torch.cudatorch.cuda..OutOfMemoryErrorOutOfMemoryError: : CUDA out of memory. Tried to allocate 2.00 GiB (GPU 0; 79.15 GiB total capacity; 77.13 GiB already allocated; 307.31 MiB free; 78.21 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONFCUDA out of memory. Tried to allocate 2.00 GiB (GPU 3; 79.15 GiB total capacity; 77.13 GiB already allocated; 355.31 MiB free; 78.21 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF
I'm confused about this situation, lora should be a peft method, why this OOM issue could occur when applying 4xA100(80GB)?
What could be the reason, please?
Hi @YJiangcm, may I ask is the codes for parallel data construction is included in this repository? If not, could you add that code please? Thank you so much.
Hello👋
Thank you for your great work for the community!
I've been trying to reproduce your results, but I'm a bit stuck. Due to the limited computing resources, I can only train Llama2-7b using LoRA. Still, there is only an evaluation script for full FT. So, I'm not sure how to use this script to evaluate the LoRA model.
Thank you in advance for your time and support!
Hi @YJiangcm, the performance of the model trained using the code in this repository differs from that claimed in the paper.
At your convenience, would you provide the trained models for evaluation please? Thank you so much.
The base method adopted by LTE is the IKE method, but the batch_editor file shows that the IKE method does not support batch edit, so how is the batch edit experiment carried out in the paper?
Hi there,
I am running the code in the repo and notice that prompt
and target_new
are being used directly as updated information
without retrieval. I'm curious about how you've approached the ablation study for the retrieval number k. Are you following the IKE method? Also, would it be possible to share more code to shed light on this process?
Thanks for the fantastic work you're doing! I can't wait to hear back from you.
A declarative, efficient, and flexible JavaScript library for building user interfaces.
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
The Web framework for perfectionists with deadlines.
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
We are working to build community through open source technology. NB: members must have two-factor auth.
Open source projects and samples from Microsoft.
Google ❤️ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.