DocumentationFine-tune ModelsModels

Model Catalog

View as Markdown

Explore the model families and sizes supported by NVIDIA NeMo Customizer.

For information on setting up model entities for customization, see the Manage Model Entities guide. For fine-tuning and deployment tutorials, see the Tutorials guide.

Before You Start

If downloading models hosted on Hugging Face, create a secret with your HuggingFace API key, then create a FileSet and Model Entity referencing the model. See index for setup instructions.

Model Families

Llama Models

View the available Llama models from Meta, ranging from 8 billion to 70 billion parameters.

Llama Nemotron Models

View the available Llama Nemotron models from NVIDIA, including Nano and Super variants for efficient and advanced instruction tuning.

Phi Models

View the available Phi models from Microsoft, designed for strong reasoning capabilities with efficient deployment.

Embedding Models

View the available embedding models optimized for retrieval and question-answering tasks.

GPT-OSS Models

View the available GPT-OSS models supported for customization.

Qwen Models

View the available Qwen models from Alibaba Cloud, including compact variants for efficient customization.

Mistral Models

View the available Mistral models, including Mistral and Ministral variants for instruction-following and reasoning tasks.

Tested Models

The following table lists models that NVIDIA tested and their available features. This is a list of known-good combinations, not a list of limits: NeMo Customizer can fine-tune many models and regimes beyond those listed, including additional Hugging Face checkpoints, other fine-tuning regimes (LoRA, merged-LoRA, full-weight, distillation), and either training backend (Automodel or Unsloth). Models and regimes outside this table may work but have not been formally validated.

For detailed technical specifications of each model such as architecture, parameters, and token limits, refer to the model family pages.

Large Language Models

The following models support both chat and completion model training.

ModelTrain a Chat Model with Tool CallingFine-tuning OptionsSequence Packing1Inference with NIMReasoning
meta-llama/Llama-3.2-3B-InstructYesFull SFT, LoRAYesSupportedNo
meta-llama/Llama-3.2-1B-InstructYesFull SFT, LoRAYesSupportedNo
meta-llama/Llama-3.1-8B-InstructYesFull SFT, LoRAYesSupportedNo
nvidia/Llama-3.1-Nemotron-Nano-8B-v1NoFull SFT, LoRAYesSupportedYes
nvidia/NVIDIA-Nemotron-Nano-9B-v2NoFull SFT, LoRANoSupportedYes
nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16NoFull SFT, LoRANoSupported (only Full SFT)Yes
nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16NoLoRANoSupportedYes
microsoft/phi-4NoFull SFT, LoRANoSupportedNo
openai/gpt-oss-20bYesFull SFT, LoRANoSupportedYes
Qwen/Qwen2.5-1.5B-InstructNoFull SFT, LoRANoSupportedYes
Qwen/Qwen3-0.6BNoFull SFT, LoRANoSupportedYes
mistralai/Mistral-7B-Instruct-v0.3NoFull SFT, LoRANoSupportedNo
mistralai/Ministral-3-3B-Instruct-2512NoFull SFT, LoRANoNoNo
mistralai/Ministral-3-3B-Reasoning-2512NoFull SFT, LoRAYesNoYes

Embedding Models

ModelFine-tuning OptionsInference with NIM
nvidia/llama-nemotron-embed-1b-v2Full SFT, LoRA (merged)Supported

For detailed technical specifications and configuration information for embedding models, see the Embedding Models page.

Footnotes

  1. Read more on sequence packing with NeMo Framework ↩