MiMo-V2-Flash#
MiMo-V2-Flash is Xiaomi’s
hybrid attention Mixture-of-Experts language model. It alternates full and
sliding-window attention layers, uses a sigmoid_with_bias router with
group-limited expert routing, and ships as an FP8 HF checkpoint.
Task |
Text Generation (MoE, hybrid attention) |
Architecture |
|
Parameters |
Approx. several hundred B total / much smaller active |
HF Org |
Available Models#
MiMo-V2-Flash: hybrid full/sliding-window attention with FP8 weights.
Architecture#
MiMoV2FlashForCausalLMSliding-window attention via the
MiMoV2FlashAttention(is_swa=True)path.MoE blocks use the shared
nemo_automodel.components.moe.layers.MoEwithscore_func="sigmoid_with_bias"andgate_precision=fp32so routing decisions stay numerically stable when activations are bf16.FP8 round-trip in
MiMoV2FlashStateDictAdaptercovers the bulk of attention/expert weights; layer norms, the gate,lm_head, andembed_tokensstay in bf16 perNON_QUANTIZED_KEY_PATTERNS.
Example HF Models#
Model |
HF ID |
|---|---|
MiMo-V2-Flash |
Example Recipes#
Recipe |
Description |
|---|---|
SFT — MiMo-V2-Flash on HellaSwag |
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/llm_finetune/mimo_v2_flash/mimo_v2_flash_hellaswag.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:
cd /opt/Automodel
3. Run the recipe:
automodel --nproc-per-node=8 examples/llm_finetune/mimo_v2_flash/mimo_v2_flash_hellaswag.yaml
See the Installation Guide and LLM Fine-Tuning Guide.
Fine-Tuning#
See the LLM Fine-Tuning Guide.