Kimi-K2.5 | NVIDIA NeMo AutoModel

Kimi-K2.5 has a checked-in NeMo AutoModel recipe for image-text-to-text. The Hugging Face configuration declares the KimiK25ForConditionalGeneration architecture.

Fine-Tune Kimi-K2.5

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

automodel examples/vlm_finetune/kimi/kimi25vl_medpix.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.
Try another checked-in recipe for this model	Use the alternate recipe.

Configuration

Setting	Configuration
Hardware	-
Strategy	FSDP2; tp_size=1, pp_size=8, cp_size=1, ep_size=32
Nodes	32
Features	Transformer Engine, DeepEP (dispatcher=deepep)
Advanced	bfloat16, attention=te, linear=torch, experts=gmm, dispatcher=deepep

More Recipes for This Model

kimi25vl_lora.yaml

Model Reference

Model Architecture

Property	Value
Task	Image-text-to-text
Architecture	`KimiK25ForConditionalGeneration`

Available Models

moonshotai/Kimi-K2.5

Vision-language model fine-tuning guide