Kimi-VL#
Kimi-VL and Kimi-K25-VL are vision language models from Moonshot AI. Kimi-VL-A3B uses a MoE language backbone (3B active parameters) with a vision encoder, supporting image understanding and multimodal reasoning.
Task |
Image-Text-to-Text |
Architecture |
|
Parameters |
~3B active (MoE) |
HF Org |
Available Models#
Kimi-VL-A3B-Instruct
Kimi-K25-VL
Architecture#
KimiVLForConditionalGeneration
Example HF Models#
Model |
HF ID |
|---|---|
Kimi-VL-A3B-Instruct |
Example Recipes#
Recipe |
Dataset |
Description |
|---|---|---|
cord-v2 |
SFT — Kimi-VL on CORD-v2 |
|
MedPix-VQA |
SFT — Kimi-K25-VL on MedPix |
Try with NeMo AutoModel#
1. Install (full instructions):
pip install nemo-automodel
2. Clone the repo to get the example recipes:
git clone https://github.com/NVIDIA-NeMo/Automodel.git
cd Automodel
3. Run the recipe from inside the repo:
automodel --nproc-per-node=8 examples/vlm_finetune/kimi/kimi2vl_cordv2.yaml
Run with Docker
1. Pull the container and mount a checkpoint directory:
docker run --gpus all -it --rm \
--shm-size=8g \
-v $(pwd)/checkpoints:/opt/Automodel/checkpoints \
nvcr.io/nvidia/nemo-automodel:26.02.00
2. Navigate to the AutoModel directory (where the recipes are):
cd /opt/Automodel
3. Run the recipe:
automodel --nproc-per-node=8 examples/vlm_finetune/kimi/kimi2vl_cordv2.yaml
See the Installation Guide and VLM Fine-Tuning Guide.
Fine-Tuning#
See the VLM Fine-Tuning Guide.