DeepSeek-V3.2 | NVIDIA NeMo AutoModel

DeepSeek-V3.2 has a checked-in NeMo AutoModel recipe for text generation. The Hugging Face configuration declares the DeepseekV32ForCausalLM architecture.

Fine-Tune DeepSeek-V3.2

Follow the installation instructions, then run the recipe from the repository root:

automodel examples/llm_benchmark/deepseek/dsv32_lora.yaml --nproc-per-node 8

Use the Slurm launcher guide for the multi-node run.

Choose a Workflow

Goal	Start Here
Run the primary recipe for this model	Use the recipe configuration.
Prepare your environment	Follow the installation instructions.

Configuration

Setting	Configuration
Hardware	-
Strategy	FSDP2; tp_size=1, pp_size=4, cp_size=1, ep_size=32
Nodes	16
Features	Activation checkpointing (activation_checkpointing=true), LoRA, Transformer Engine, DeepEP (dispatcher=deepep)
Advanced	attention=te, linear=te, experts=gmm, dispatcher=deepep

Model Reference

Model Architecture

Property	Value
Task	Text generation
Architecture	`DeepseekV32ForCausalLM`

Available Models

deepseek-ai/DeepSeek-V3.2

LLM fine-tuning guide