MiniCPM

MiniCPM is a compact language model series from OpenBMB and Tsinghua University, designed to deliver strong performance at small parameter counts using model merging and continuous training techniques.


Task	Text Generation
Architecture	`MiniCPMForCausalLM` / `MiniCPM3ForCausalLM` / `MiniCPM5ForCausalLM`
Parameters	1B – 4B
HF Org	openbmb

Available Models

MiniCPM5-1B (MiniCPM5ForCausalLM): 1B
MiniCPM3-4B (MiniCPM3ForCausalLM): 4B
MiniCPM-2B-sft-bf16 (MiniCPMForCausalLM): 2B, SFT
MiniCPM-2B-dpo-bf16 (MiniCPMForCausalLM): 2B, DPO

Architectures

MiniCPMForCausalLM — MiniCPM v1/v2
MiniCPM3ForCausalLM — MiniCPM3
MiniCPM5ForCausalLM — MiniCPM5

Example HF Models

Model	HF ID
MiniCPM5 1B	`openbmb/MiniCPM5-1B`
MiniCPM 2B SFT	`openbmb/MiniCPM-2B-sft-bf16`
MiniCPM3 4B	`openbmb/MiniCPM3-4B`

Try with NeMo AutoModel

Install NeMo AutoModel and follow the fine-tuning guide to configure a recipe for this model.

1. Install (full instructions):

$ pip install nemo-automodel

2. Clone the repo to get example recipes you can adapt:

$ git clone https://github.com/NVIDIA-NeMo/Automodel.git
$ cd Automodel

3. Fine-tune using one of the provided MiniCPM5 recipes:

$ # Full fine-tuning
$ automodel examples/llm_finetune/minicpm/minicpm5_1b_squad.yaml --nproc-per-node 8
$ 
$ # LoRA fine-tuning
$ automodel examples/llm_finetune/minicpm/minicpm5_1b_squad_peft.yaml --nproc-per-node 8

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:26.06.00

2. The recipes are at /opt/Automodel/examples/ — navigate there:

$ cd /opt/Automodel

3. Fine-tune:

$ automodel --nproc-per-node=8 examples/llm_finetune/minicpm/minicpm5_1b_squad.yaml \
>   --model.pretrained_model_name_or_path <MODEL_HF_ID>

Refer to the Installation Guide and LLM Fine-Tuning Guide.

Fine-Tune

Refer to the LLM Fine-Tuning Guide.