When Scaling Meets LLM Finetuning: The Effect of Data, Model and Finetuning Method
Paper • 2402.17193 • Published • 27
Fine tuning
-
-
What Happened in LLMs Layers when Trained for Fast vs. Slow Thinking: A Gradient Perspective
Paper • 2410.23743 • Published • 64 -
Direct Preference Optimization Using Sparse Feature-Level Constraints
Paper • 2411.07618 • Published • 17 -
Transformer^2: Self-adaptive LLMs
Paper • 2501.06252 • Published • 55 -
Control LLM: Controlled Evolution for Intelligence Retention in LLM
Paper • 2501.10979 • Published • 6 -
Taming LLMs by Scaling Learning Rates with Gradient Grouping
Paper • 2506.01049 • Published • 40 -
Leveraging Self-Attention for Input-Dependent Soft Prompting in LLMs
Paper • 2506.05629 • Published • 37 -
All is Not Lost: LLM Recovery without Checkpoints
Paper • 2506.15461 • Published • 41 -
SRFT: A Single-Stage Method with Supervised and Reinforcement Fine-Tuning for Reasoning
Paper • 2506.19767 • Published • 15 -
Optimizing ML Training with Metagradient Descent
Paper • 2503.13751 • Published • 1 -
Towards a Unified View of Large Language Model Post-Training
Paper • 2509.04419 • Published • 77 -
Analyzing the Effects of Supervised Fine-Tuning on Model Knowledge from Token and Parameter Levels
Paper • 2509.16596 • Published • 14 -
Fine-tuning Done Right in Model Editing
Paper • 2509.22072 • Published • 28 -
Interactive Training: Feedback-Driven Neural Network Optimization
Paper • 2510.02297 • Published • 43 -
LightMem: Lightweight and Efficient Memory-Augmented Generation
Paper • 2510.18866 • Published • 116 -
π_RL: Online RL Fine-tuning for Flow-based Vision-Language-Action Models
Paper • 2510.25889 • Published • 66 -
ROOT: Robust Orthogonalized Optimizer for Neural Network Training
Paper • 2511.20626 • Published • 45 -
Good SFT Optimizes for SFT, Better SFT Prepares for Reinforcement Learning
Paper • 2602.01058 • Published • 45 -
MSign: An Optimizer Preventing Training Instability in Large Language Models via Stable Rank Restoration
Paper • 2602.01734 • Published • 34 -
Unified Latents (UL): How to train your latents
Paper • 2602.17270 • Published • 62 -
MegaTrain: Full Precision Training of 100B+ Parameter Large Language Models on a Single GPU
Paper • 2604.05091 • Published • 47 -
How to Fine-Tune a Reasoning Model? A Teacher-Student Cooperation Framework to Synthesize Student-Consistent SFT Data
Paper • 2604.14164 • Published • 35 -
GFT: From Imitation to Reward Fine-Tuning with Unbiased Group Advantages and Dynamic Coefficient Rectification
Paper • 2604.14258 • Published • 23 -
Context Unrolling in Omni Models
Paper • 2604.21921 • Published • 14 -
Efficient Training on Multiple Consumer GPUs with RoundPipe
Paper • 2604.27085 • Published • 47 -
Geometry Conflict: Explaining and Controlling Forgetting in LLM Continual Post-Training
Paper • 2605.09608 • Published • 53 -
It Takes Two: Complementary Self-Distillation for Contextual Integrity in LLMs
Paper • 2605.20258 • Published • 30 -
HRM-Text: Efficient Pretraining Beyond Scaling
Paper • 2605.20613 • Published • 323 -
How LoRA Remembers? A Parametric Memory Law for LLM Finetuning
Paper • 2605.30260 • Published • 44 -
Which Pretraining Paradigm Better Serves Spatial Intelligence? An Empirical Comparison of Vision-Language and Video Generation Models
Paper • 2605.28132 • Published • 25 -
On the Scaling of PEFT: Towards Million Personal Models of Trillion Parameters
Paper • 2606.02437 • Published • 243 -
NITP: Next Implicit Token Prediction for LLM Pre-training
Paper • 2605.24956 • Published • 36 -
Code2LoRA: Hypernetwork-Generated Adapters for Code Language Models under Software Evolution
Paper • 2606.06492 • Published • 96 -
LatentSkill: From In-Context Textual Skills to In-Weight Latent Skills for LLM Agents
Paper • 2606.06087 • Published • 67 -
Skip a Layer or Loop It? Learning Program-of-Layers in LLMs
Paper • 2606.06574 • Published • 25 -
Towards Mechanistically Understanding Why Memorized Knowledge Fails to Generalize in Large Language Model Finetuning
Paper • 2607.08393 • Published • 19 -
SLAI T-Rex: Full-Parameter Post-training of the DeepSeek-V4 Family on Ascend SuperPOD
Paper • 2607.20145 • Published • 74 -
Agentic ESOpt: Fine-Tuning Long-Horizon LLM Agents with Minimal GPU Requirements
Paper • 2608.17310 • Published • 96