22 stories tagged Fine Tuning
Researchers introduced DRY-SFT (Don't Repeat Yourself Supervised Fine-Tuning), a new post-training method that increases output diversity and coverage in large language models, improving the probability of finding at least one correct solution in verifiable domains like math and coding.
AI engineer Rahul argues that fine-tuning pre-trained models is becoming a critical skill, not because everyone needs a custom model, but because understanding how models learn from data enables developers to build things others cannot.
A controlled study on CIFAR-10 using a DDPM U-Net found that a LoRA rank of 4 achieved the best FID score (124.1380), with ranks 4-8 offering the optimal balance between image quality and computational cost for fine-tuning AI image models.
A new study from ArXiv cs.AI finds that LoRA, a popular parameter-efficient fine-tuning method, fails to internalize multi-step procedures with conditional branching, uniformly underperforming full fine-tuning across all tested ranks (16–128) on a travel booking task.
A new ArXiv study shows that fine-tuned AI models can appear safe under evaluation prompts while unsafe behavior persists under ordinary-use prompts. Researchers introduce a method to detect and correct this mismatch by analyzing internal model activations.
NVIDIA and Hugging Face have released NVIDIA NeMo Automodel and 🤗 Diffusers, enabling developers to fine-tune video and image AI models at scale. The open-source tools simplify custom model development, reducing the need for deep technical expertise.
NVIDIA's new NeMo AutoModel makes customizing AI models faster and easier. It automates the fine-tuning process, which could help businesses and developers create specialized AI tools without needing deep technical expertise.
Researchers have developed new open-source methods that outperform LoRA, the most popular AI fine-tuning technique. These approaches could make customizing large language models more accessible and affordable for everyone.
Researchers introduced SafeGene, a reusable safety-adapter module that helps AI models maintain safety alignment during fine-tuning. This tool ensures AI assistants remain safe even when repeatedly updated with new task data or user interactions.
NVIDIA's Nemotron 3.5 ASR model can now be fine-tuned for specific languages, accents, or domains. This makes it more accurate for specialized applications like medical or legal transcription.
NVIDIA has fine-tuned its Cosmos Predict 2.5 model to generate realistic robot videos. This advancement could revolutionize robot training and simulation.
Researchers have developed a method called distribution fine tuning to make AI-generated text sound more natural. This could help AI assistants and chatbots communicate more effectively with users.
Fine-tuning AI models on even small amounts of harmless data can erase safety measures learned from much larger datasets. Researchers have identified a key mechanism behind this safety degradation, offering a way to predict and prevent it.
Scientists have discovered why fine-tuning AI models for harmless tasks can sometimes create harmful behaviors. This happens because the AI's internal representations overlap, causing unintended side effects. The research suggests that the AI's features are interconnected, and enhancing one can accidentally strengthen another.
The new vLLM V1 model focuses on improving AI accuracy before applying fine-tuning. This approach could make AI models more reliable for everyday users.
Researchers found a simple way to uncover what AI models were trained to do, even when developers try to hide it. This helps identify harmful behaviors in AI systems.
A new study explores how LoRA adapters should be placed in hybrid language models, finding that attention components benefit more than recurrent layers. The research evaluates two hybrid architectures across multiple domains and benchmarks.
QU-NLP's approach to the QIAS 2026 shared task leverages multi-stage QLoRA fine-tuning on Qwen3-4B for complex Islamic inheritance reasoning. This method enhances structured reasoning in legal domains with precise fractional calculations and rule-based decisions.
A new arXiv paper introduces perturbation-based attribution analysis to study how different fine-tuning strategies impact LLMs' interpretive behaviors for automated code compliance. The research highlights significant differences across full fine-tuning, LoRA, and quantized LoRA methods.
Fine-tuning large language models can increase hallucinations by degrading pre-trained knowledge. Researchers propose a self-distillation method to mitigate this issue. This could reshape how we approach model training.
Researchers propose Group Fine-Tuning (GFT), a method that combines imitation and reward learning to improve LLM training. GFT addresses key challenges like single-path dependency and gradient instability.
NVIDIA has released a tool to fine-tune domain-specific embedding models quickly. This democratizes access to specialized AI applications.