NVIDIA released LoRA fine-tuning recipes for biological foundation models (BioNeMo). The composition is strongly academic—one official NVIDIA source plus four arXiv papers—so it reads less as an announcement than research accumulation bundled into an implementation guide. The direction is adapting huge biology models with LoRA rather than retraining them, targeting specific drug-discovery and protein tasks; the through-line is a move from 'train big' to 'adapt light.' The notable part is domain-specific adaptation methods organized for practitioners in life sciences. It's still research-led; real-world reproducibility and adoption aren't visible right after release.
NVIDIA × Training & Fine-tuning
NVIDIA ships BioNeMo LoRA recipes
NVIDIA ships BioNeMo LoRA recipes
✎ Story body
▲ Official & Press
Official
Fine-Tuning Biological Foundation Models with LoRA Using NVIDIA BioNeMo Recipes
NVIDIA details LoRA fine-tuning of biological foundation models via BioNeMo
Academic (arxiv etc.) 31 ▾
Academic
Native Active Perception as Reasoning for Omni-Modal Understanding
Active perception as reasoning for efficient omni-modal understanding
Academic
Measuring commonsense and knowledge retention in VLA models
Academic
Trade-offs in Medical LLM Adaptation: An Empirical Study in French QA
Trade-offs in medical LLM adaptation, studied on French QA
Academic
Mechanism-Guided Selective Unlearning for RLVR-Induced Reasoning
MAST selectively unlearns RLVR-induced reasoning with less damage
Academic
Beyond Safe Data: Pretraining-Stage Alignment with Regular Safety Reflection
Pretraining-stage alignment via regular safety reflection
Academic
ProductConsistency preserves product identity in instruction-based editing
Academic
ARIADNE: Agnostic Routing for Inference-time Adapter DyNamic sElection
ARIADNE: agnostic routing for inference-time adapter selection
Academic
Learning from the Self-future: On-policy Self-distillation for dLLMs
On-policy self-distillation explored for diffusion LLMs
Academic
Multi-Source Cybersecurity Logs: An ATT&CK-Labeled Dataset and SLM Evaluation
ATT&CK-labeled multi-source security log dataset with SLM evaluation
Academic
S4oP: Operator-level Pruning of Structured State Space Models for Resource-Constrained Devices
S4oP prunes structured state space models at the operator level
Academic
EAGG: Embodiment-Aligned Grasp Generation via Geometry-Aware Graph Conditioning
EAGG: embodiment-aligned grasp generation via graph conditioning
Academic
From reasoning traces to reusable modules for compositional reasoning
Academic
Uncertainty Quantification for Flow-Based Vision-Language-Action Models
Uncertainty quantification for flow-based vision-language-action models
Academic
When English Isn't the Best Teacher: Source Language Effects in Cross-Lingual In-Context Learning
Source-language effects in cross-lingual in-context learning
Academic
Catastrophic Forgetting is Low-Rank: A Function-Space Theory for Continual Adaptation
Catastrophic forgetting is low-rank: a function-space theory
Academic
Fine-tuning LLMs for Passive Depression Severity Estimation from AI Mental Health Dialogue
Fine-tuning LLMs for passive depression severity from AI dialogue
Academic
Environment-Grounded Automated Prompt Optimization for LLM Game Agents
Environment-grounded automated prompt optimization for LLM game agents
Academic
From Drift to Coherence: Stabilizing Beliefs in LLMs
From drift to coherence: stabilizing beliefs in LLMs
Academic
Improving low-resource ASR via bilingual fine-tuning with language ID
Academic
SuCo: Sufficiency-guided Continuous Adaptive Reasoning
SuCo: sufficiency-guided continuous adaptive reasoning
Academic
The Value Axis: Language Models Encode Whether They're on the Right Track
LLMs encode a 'value axis' tracking if their strategy works
Academic
Hierarchical Advantage Weighting for Online RL Fine-Tuning of VLAs from Sparse Episode Outcomes
HABC: hierarchical advantage weighting for RL fine-tuning of VLAs
Academic
KVEraser: Learning to Steer KV Cache for Efficient Localized Context Erasing
KVEraser edits the KV cache to erase context efficiently
Academic
ExpRL: Exploratory RL for LLM Mid-Training
ExpRL uses human QA as reward scaffolds for LLM mid-training RL
Academic
Measurement study of post-hoc falsification operators for code models
Academic
Exploring Extrinsic and Intrinsic Properties for Effective Reasoning with Code Interpreter
Study probes extrinsic and intrinsic traits of code-interpreter reasoning
Academic
RDS Fusion: neuro-symbolic gating with compressed CoT for irony detection
Academic
ATOM-Bench evaluates atomic skills and compositional generalization in robots
Academic
Adaptive and Explicit safe: Triggering Latent Safety Awareness in Large Reasoning Models
Triggering latent safety awareness to harden large reasoning models
Academic
MIXGUARD: mixup-based privacy for LLM split learning
Academic
daVinci-kernel: an RL framework co-evolving skills for GPU kernel tuning