ShiZhengyan/InstructionModelling
[NeurIPS 2024 Main Track] Code for the paper titled "Instruction Tuning With Loss Over Instructions"
This project offers a new method to fine-tune large language models (LLMs) to better understand and follow instructions. It takes existing LLMs and instruction-output training datasets, applying a unique loss function that considers the instruction part of the data, not just the desired output. The result is a more accurate and responsive LLM, particularly useful for researchers and practitioners working on improving LLM performance on diverse natural language processing and generation tasks.
No commits in the last 6 months.
Use this if you are a researcher or AI engineer working to improve the instruction-following capabilities of language models, especially when training with datasets that have long instructions and short desired outputs, or when working with limited training data.
Not ideal if you are a non-technical end-user simply looking for a ready-to-use application, as this project requires deep technical knowledge of LLM training and infrastructure.
Stars
38
Forks
8
Language
Python
License
—
Category
Last pushed
May 24, 2024
Commits (30d)
0
Get this data via API
curl "https://pt-edge.onrender.com/api/v1/quality/transformers/ShiZhengyan/InstructionModelling"
Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.
Related models
raymin0223/fast_robust_early_exit
Fast and Robust Early-Exiting Framework for Autoregressive Language Models with Synchronized...
SALT-NLP/Adaptive-Compositional-Modules
Code for the ACL 2022 paper "Continual Sequence Generation with Adaptive Compositional Modules"
oooranz/Baby-CoThought
🍼 Baby's CoThought: Leveraging LLMs for Enhanced Reasoning in Compact Models (BabyLM Challenge)
joisino/zeh
Code for "Even GPT-5.2 Can’t Count to Five: The Case for Zero-Error Horizons in Trustworthy LLMs"
yhy1117/X-Mixup
Implementation of ICLR 2022 paper "Enhancing Cross-lingual Transfer by Manifold Mixup".