[model] Add bridge support for GLM-4.7-Flash

[jQizhang]

Hugging Face repository

https://huggingface.co/zai-org/GLM-4.7-Flash

Architecture family

MoE decoder-only (e.g. DeepSeek V2/V3, OLMoE, Qwen3-MoE, MiniMax-M2)

Required deliverables

• Model providers
• HF conversion bridge
• Unit tests (config and bridge)
• Model conversion functional tests
• Optimal pretraining recipe
• Optimal finetuning recipe
• Recipe unit tests
• Recipe functional tests
• End-to-end CI coverage

Owner if known

No response

Extra context

zai-org/GLM-4.7-Flash uses a new architecture class Glm4MoeLiteForCausalLM (model_type: glm4_moe_lite), which is not covered by the existing GLM45Bridge. Are there any plans to support this model? Thanks.

[yaoyu-33]

@jQizhang : are you interested in helping us with an implementation? We can help to guide you through the process. Internally, we have higher priority models need to be added first.

[snowmanwwg]

@jQizhang GLM4.7-flash is supported via https://github.com/NVIDIA-NeMo/Automodel/blob/main/examples/llm_finetune/glm/glm_4.7_flash_te_packed_sequence.yaml