GLM-4 MoE (GLM-4.5 / GLM-4.7)

GLM-4.5 and GLM-4.7 are Mixture-of-Experts variants of the GLM family released under the zai-org HuggingFace organization. GLM-4.7-Flash is a lighter variant with fewer active parameters.


Task	Text Generation (MoE)
Architecture	`Glm4MoeForCausalLM` / `Glm4MoeLiteForCausalLM`
Parameters	varies
HF Org	zai-org

Available Models

GLM-4.5-Air (Glm4MoeForCausalLM)
GLM-4.7 (Glm4MoeForCausalLM)
GLM-4.7-Flash (Glm4MoeLiteForCausalLM): lightweight MoE variant

Architectures

Glm4MoeForCausalLM — GLM-4.5, GLM-4.7
Glm4MoeLiteForCausalLM — GLM-4.7-Flash

Example HF Models

Model	HF ID
GLM-4.5-Air	`zai-org/GLM-4.5-Air`
GLM-4.7	`zai-org/GLM-4.7`
GLM-4.7-Flash	`zai-org/GLM-4.7-Flash`

Example Recipes

Recipe	Description
glm_4.5_air_te_deepep.yaml	SFT — GLM-4.5-Air with TE + DeepEP
glm_4.7_te_deepep.yaml	SFT — GLM-4.7 with TE + DeepEP
glm_4.7_flash_te_deepep.yaml	SFT — GLM-4.7-Flash with TE + DeepEP
glm_4.7_flash_te_packed_sequence.yaml	SFT — GLM-4.7-Flash with packed sequences

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

This recipe was validated on 8 nodes × 8 GPUs (64 H100s). See the Launcher Guide for multi-node setup.

3. Run the recipe from inside the repo:

$ automodel --nproc-per-node=8 examples/llm_finetune/glm/glm_4.5_air_te_deepep.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:26.06.00

2. Navigate to the AutoModel directory (where the recipes are):

$ cd /opt/Automodel

3. Run the recipe:

$ automodel --nproc-per-node=8 examples/llm_finetune/glm/glm_4.5_air_te_deepep.yaml

See the Installation Guide and LLM Fine-Tuning Guide.

Fine-Tuning

See the LLM Fine-Tuning Guide and the Large MoE Fine-Tuning Guide.